add gradio components

app.py

@@ -1,5 +1,6 @@
 import os
 import argparse
+import gradio as gr
 
 from modules.whisper.whisper_Inference import WhisperInference
 from modules.whisper.faster_whisper_inference import FasterWhisperInference
@@ -84,7 +85,7 @@ class App:
                     with gr.Column():
                         input_file = gr.Files(type="filepath", label="Upload File here")
                         tb_input_folder = gr.Textbox(label="Input Folder Path (Optional)",
-                                                     info="Optional: Specify the folder path where the input files are located, if you prefer to use local files instead of uploading them." 
+                                                     info="Optional: Specify the folder path where the input files are located, if you prefer to use local files instead of uploading them."
                                                           " Leave this field empty if you do not wish to use a local path.",
                                                      visible=self.args.colab,
                                                      value="")
@@ -97,32 +98,83 @@ class App:
                     with gr.Row():
                         cb_translate = gr.Checkbox(value=False, label="Translate to English?", interactive=True)
                     with gr.Row():
-                        cb_timestamp = gr.Checkbox(value=True, label="Add a timestamp to the end of the filename", interactive=True)
+                        cb_timestamp = gr.Checkbox(value=True, label="Add a timestamp to the end of the filename",
+                                                   interactive=True)
                     with gr.Accordion("Advanced Parameters", open=False):
                         nb_beam_size = gr.Number(label="Beam Size", value=1, precision=0, interactive=True)
-                        nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0, interactive=True)
+                        nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0,
+                                                          interactive=True)
                         nb_no_speech_threshold = gr.Number(label="No Speech Threshold", value=0.6, interactive=True)
-                        dd_compute_type = gr.Dropdown(label="Compute Type", choices=self.whisper_inf.available_compute_types, value=self.whisper_inf.current_compute_type, interactive=True)
+                        dd_compute_type = gr.Dropdown(label="Compute Type",
+                                                      choices=self.whisper_inf.available_compute_types,
+                                                      value=self.whisper_inf.current_compute_type, interactive=True)
                         nb_best_of = gr.Number(label="Best Of", value=5, interactive=True)
                         nb_patience = gr.Number(label="Patience", value=1, interactive=True)
-                        cb_condition_on_previous_text = gr.Checkbox(label="Condition On Previous Text", value=True, interactive=True)
+                        cb_condition_on_previous_text = gr.Checkbox(label="Condition On Previous Text", value=True,
+                                                                    interactive=True)
                         tb_initial_prompt = gr.Textbox(label="Initial Prompt", value=None, interactive=True)
-                        sd_temperature = gr.Slider(label="Temperature", value=0, step=0.01, maximum=1.0, interactive=True)
-                        nb_compression_ratio_threshold = gr.Number(label="Compression Ratio Threshold", value=2.4, interactive=True)
+                        sd_temperature = gr.Slider(label="Temperature", value=0, step=0.01, maximum=1.0,
+                                                   interactive=True)
+                        nb_compression_ratio_threshold = gr.Number(label="Compression Ratio Threshold", value=2.4,
+                                                                   interactive=True)
+                        with gr.Group(visible=isinstance(self.whisper_inf, FasterWhisperInference)):
+                            with gr.Column():
+                                nb_length_penalty = gr.Number(label="Length Penalty", value=1,
+                                                              info="Exponential length penalty constant.")
+                                nb_repetition_penalty = gr.Number(label="Repetition Penalty", value=1,
+                                                                  info="Penalty applied to the score of previously generated tokens (set > 1 to penalize).")
+                                nb_no_repeat_ngram_size = gr.Number(label="No Repeat N-gram Size", value=0, precision=0,
+                                                                    info="Prevent repetitions of n-grams with this size (set 0 to disable).")
+                                tb_prefix = gr.Textbox(label="Prefix", value="",
+                                                       info="Optional text to provide as a prefix for the first window.")
+                                cb_suppress_blank = gr.Checkbox(label="Suppress Blank", value=True,
+                                                                info="Suppress blank outputs at the beginning of the sampling.")
+                                tb_suppress_tokens = gr.Textbox(label="Suppress Tokens", value="-1",
+                                                                info="List of token IDs to suppress. -1 will suppress a default set of symbols as defined in the model config.json file.")
+                                nb_max_initial_timestamp = gr.Number(label="Max Initial Timestamp", value=1.0,
+                                                                     info="The initial timestamp cannot be later than this.")
+                                cb_word_timestamps = gr.Checkbox(label="Word Timestamps", value=False,
+                                                                 info="Extract word-level timestamps using the cross-attention pattern and dynamic time warping, and include the timestamps for each word in each segment.")
+                                tb_prepend_punctuations = gr.Textbox(label="Prepend Punctuations", value="\"'“¿([{-",
+                                                                     info="If word_timestamps is True, merge these punctuation symbols with the next word.")
+                                tb_append_punctuations = gr.Textbox(label="Append Punctuations",
+                                                                    value="\"'.。,，!！?？:：”)]}、",
+                                                                    info="If word_timestamps is True, merge these punctuation symbols with the previous word.")
+                                nb_max_new_tokens = gr.Number(label="Max New Tokens", value=None, precision=0,
+                                                              info="Maximum number of new tokens to generate per-chunk. If not set, the maximum will be set by the default max_length.")
+                                nb_chunk_length = gr.Number(label="Chunk Length", value=None,
+                                                            info="The length of audio segments. If it is not None, it will overwrite the default chunk_length of the FeatureExtractor.")
+                                nb_hallucination_silence_threshold = gr.Number(label="Hallucination Silence Threshold",
+                                                                               value=None,
+                                                                               info="When word_timestamps is True, skip silent periods longer than this threshold (in seconds) when a possible hallucination is detected.")
+                                tb_hotwords = gr.Textbox(label="Hotwords", value="",
+                                                         info="Hotwords/hint phrases to provide the model with. Has no effect if prefix is not None.")
+                                nb_language_detection_threshold = gr.Number(label="Language Detection Threshold",
+                                                                            value=None,
+                                                                            info="If the maximum probability of the language tokens is higher than this value, the language is detected.")
+                                nb_language_detection_segments = gr.Number(label="Language Detection Segments", value=1,
+                                                                           precision=0,
+                                                                           info="Number of segments to consider for the language detection.")
                     with gr.Accordion("VAD", open=False):
                         cb_vad_filter = gr.Checkbox(label="Enable Silero VAD Filter", value=False, interactive=True)
-                        sd_threshold = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label="Speech Threshold", value=0.5, info="Lower it to be more sensitive to small sounds.")
-                        nb_min_speech_duration_ms = gr.Number(label="Minimum Speech Duration (ms)", precision=0, value=250)
+                        sd_threshold = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label="Speech Threshold",
+                                                 value=0.5, info="Lower it to be more sensitive to small sounds.")
+                        nb_min_speech_duration_ms = gr.Number(label="Minimum Speech Duration (ms)", precision=0,
+                                                              value=250)
                         nb_max_speech_duration_s = gr.Number(label="Maximum Speech Duration (s)", value=9999)
-                        nb_min_silence_duration_ms = gr.Number(label="Minimum Silence Duration (ms)", precision=0, value=2000)
+                        nb_min_silence_duration_ms = gr.Number(label="Minimum Silence Duration (ms)", precision=0,
+                                                               value=2000)
                         nb_speech_pad_ms = gr.Number(label="Speech Padding (ms)", precision=0, value=400)
                     with gr.Accordion("Diarization", open=False):
                         cb_diarize = gr.Checkbox(label="Enable Diarization")
                         tb_hf_token = gr.Text(label="HuggingFace Token", value="",
                                               info="This is only needed the first time you download the model. If you already have models, you don't need to enter. "
-                                                    "To download the model, you must manually go to \"https://huggingface.co/pyannote/speaker-diarization-3.1\" and agree to their requirement.")
-                        dd_diarization_device = gr.Dropdown(label="Device", choices=self.whisper_inf.diarizer.get_available_device(), value=self.whisper_inf.diarizer.get_device())
-                    with gr.Accordion("Insanely Fast Whisper Parameters", open=False, visible=isinstance(self.whisper_inf, InsanelyFastWhisperInference)):
+                                                   "To download the model, you must manually go to \"https://huggingface.co/pyannote/speaker-diarization-3.1\" and agree to their requirement.")
+                        dd_diarization_device = gr.Dropdown(label="Device",
+                                                            choices=self.whisper_inf.diarizer.get_available_device(),
+                                                            value=self.whisper_inf.diarizer.get_device())
+                    with gr.Accordion("Insanely Fast Whisper Parameters", open=False,
+                                      visible=isinstance(self.whisper_inf, InsanelyFastWhisperInference)):
                         nb_chunk_length_s = gr.Number(label="Chunk Lengths (sec)", value=30, precision=0)
                         nb_batch_size = gr.Number(label="Batch Size", value=24, precision=0)
                     with gr.Row():
@@ -133,30 +185,48 @@ class App:
                         btn_openfolder = gr.Button('📂', scale=1)
 
                     params = [input_file, tb_input_folder, dd_file_format, cb_timestamp]
-                    whisper_params = WhisperParameters(model_size=dd_model,
-                                                       lang=dd_lang,
-                                                       is_translate=cb_translate,
-                                                       beam_size=nb_beam_size,
-                                                       log_prob_threshold=nb_log_prob_threshold,
-                                                       no_speech_threshold=nb_no_speech_threshold,
-                                                       compute_type=dd_compute_type,
-                                                       best_of=nb_best_of,
-                                                       patience=nb_patience,
-                                                       condition_on_previous_text=cb_condition_on_previous_text,
-                                                       initial_prompt=tb_initial_prompt,
-                                                       temperature=sd_temperature,
-                                                       compression_ratio_threshold=nb_compression_ratio_threshold,
-                                                       vad_filter=cb_vad_filter,
-                                                       threshold=sd_threshold,
-                                                       min_speech_duration_ms=nb_min_speech_duration_ms,
-                                                       max_speech_duration_s=nb_max_speech_duration_s,
-                                                       min_silence_duration_ms=nb_min_silence_duration_ms,
-                                                       speech_pad_ms=nb_speech_pad_ms,
-                                                       chunk_length_s=nb_chunk_length_s,
-                                                       batch_size=nb_batch_size,
-                                                       is_diarize=cb_diarize,
-                                                       hf_token=tb_hf_token,
-                                                       diarization_device=dd_diarization_device)
+                    whisper_params = WhisperParameters(
+                        model_size=dd_model,
+                        lang=dd_lang,
+                        is_translate=cb_translate,
+                        beam_size=nb_beam_size,
+                        log_prob_threshold=nb_log_prob_threshold,
+                        no_speech_threshold=nb_no_speech_threshold,
+                        compute_type=dd_compute_type,
+                        best_of=nb_best_of,
+                        patience=nb_patience,
+                        condition_on_previous_text=cb_condition_on_previous_text,
+                        initial_prompt=tb_initial_prompt,
+                        temperature=sd_temperature,
+                        compression_ratio_threshold=nb_compression_ratio_threshold,
+                        vad_filter=cb_vad_filter,
+                        threshold=sd_threshold,
+                        min_speech_duration_ms=nb_min_speech_duration_ms,
+                        max_speech_duration_s=nb_max_speech_duration_s,
+                        min_silence_duration_ms=nb_min_silence_duration_ms,
+                        speech_pad_ms=nb_speech_pad_ms,
+                        chunk_length_s=nb_chunk_length_s,
+                        batch_size=nb_batch_size,
+                        is_diarize=cb_diarize,
+                        hf_token=tb_hf_token,
+                        diarization_device=dd_diarization_device,
+                        length_penalty=nb_length_penalty,
+                        repetition_penalty=nb_repetition_penalty,
+                        no_repeat_ngram_size=nb_no_repeat_ngram_size,
+                        prefix=tb_prefix,
+                        suppress_blank=cb_suppress_blank,
+                        suppress_tokens=tb_suppress_tokens,
+                        max_initial_timestamp=nb_max_initial_timestamp,
+                        word_timestamps=cb_word_timestamps,
+                        prepend_punctuations=tb_prepend_punctuations,
+                        append_punctuations=tb_append_punctuations,
+                        max_new_tokens=nb_max_new_tokens,
+                        chunk_length=nb_chunk_length,
+                        hallucination_silence_threshold=nb_hallucination_silence_threshold,
+                        hotwords=tb_hotwords,
+                        language_detection_threshold=nb_language_detection_threshold,
+                        language_detection_segments=nb_language_detection_segments
+                    )
 
                     btn_run.click(fn=self.whisper_inf.transcribe_file,
                                   inputs=params + whisper_params.as_list(),
@@ -186,28 +256,77 @@ class App:
                                                    interactive=True)
                     with gr.Accordion("Advanced Parameters", open=False):
                         nb_beam_size = gr.Number(label="Beam Size", value=1, precision=0, interactive=True)
-                        nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0, interactive=True)
+                        nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0,
+                                                          interactive=True)
                         nb_no_speech_threshold = gr.Number(label="No Speech Threshold", value=0.6, interactive=True)
-                        dd_compute_type = gr.Dropdown(label="Compute Type", choices=self.whisper_inf.available_compute_types, value=self.whisper_inf.current_compute_type, interactive=True)
+                        dd_compute_type = gr.Dropdown(label="Compute Type",
+                                                      choices=self.whisper_inf.available_compute_types,
+                                                      value=self.whisper_inf.current_compute_type, interactive=True)
                         nb_best_of = gr.Number(label="Best Of", value=5, interactive=True)
                         nb_patience = gr.Number(label="Patience", value=1, interactive=True)
-                        cb_condition_on_previous_text = gr.Checkbox(label="Condition On Previous Text", value=True, interactive=True)
+                        cb_condition_on_previous_text = gr.Checkbox(label="Condition On Previous Text", value=True,
+                                                                    interactive=True)
                         tb_initial_prompt = gr.Textbox(label="Initial Prompt", value=None, interactive=True)
-                        sd_temperature = gr.Slider(label="Temperature", value=0, step=0.01, maximum=1.0, interactive=True)
-                        nb_compression_ratio_threshold = gr.Number(label="Compression Ratio Threshold", value=2.4, interactive=True)
+                        sd_temperature = gr.Slider(label="Temperature", value=0, step=0.01, maximum=1.0,
+                                                   interactive=True)
+                        nb_compression_ratio_threshold = gr.Number(label="Compression Ratio Threshold", value=2.4,
+                                                                   interactive=True)
+                        with gr.Group(visible=isinstance(self.whisper_inf, FasterWhisperInference)):
+                            with gr.Column():
+                                nb_length_penalty = gr.Number(label="Length Penalty", value=1,
+                                                              info="Exponential length penalty constant.")
+                                nb_repetition_penalty = gr.Number(label="Repetition Penalty", value=1,
+                                                                  info="Penalty applied to the score of previously generated tokens (set > 1 to penalize).")
+                                nb_no_repeat_ngram_size = gr.Number(label="No Repeat N-gram Size", value=0, precision=0,
+                                                                    info="Prevent repetitions of n-grams with this size (set 0 to disable).")
+                                tb_prefix = gr.Textbox(label="Prefix", value="",
+                                                       info="Optional text to provide as a prefix for the first window.")
+                                cb_suppress_blank = gr.Checkbox(label="Suppress Blank", value=True,
+                                                                info="Suppress blank outputs at the beginning of the sampling.")
+                                tb_suppress_tokens = gr.Textbox(label="Suppress Tokens", value="-1",
+                                                                info="List of token IDs to suppress. -1 will suppress a default set of symbols as defined in the model config.json file.")
+                                nb_max_initial_timestamp = gr.Number(label="Max Initial Timestamp", value=1.0,
+                                                                     info="The initial timestamp cannot be later than this.")
+                                cb_word_timestamps = gr.Checkbox(label="Word Timestamps", value=False,
+                                                                 info="Extract word-level timestamps using the cross-attention pattern and dynamic time warping, and include the timestamps for each word in each segment.")
+                                tb_prepend_punctuations = gr.Textbox(label="Prepend Punctuations", value="\"'“¿([{-",
+                                                                     info="If word_timestamps is True, merge these punctuation symbols with the next word.")
+                                tb_append_punctuations = gr.Textbox(label="Append Punctuations",
+                                                                    value="\"'.。,，!！?？:：”)]}、",
+                                                                    info="If word_timestamps is True, merge these punctuation symbols with the previous word.")
+                                nb_max_new_tokens = gr.Number(label="Max New Tokens", value=None, precision=0,
+                                                              info="Maximum number of new tokens to generate per-chunk. If not set, the maximum will be set by the default max_length.")
+                                nb_chunk_length = gr.Number(label="Chunk Length", value=None,
+                                                            info="The length of audio segments. If it is not None, it will overwrite the default chunk_length of the FeatureExtractor.")
+                                nb_hallucination_silence_threshold = gr.Number(label="Hallucination Silence Threshold",
+                                                                               value=None,
+                                                                               info="When word_timestamps is True, skip silent periods longer than this threshold (in seconds) when a possible hallucination is detected.")
+                                tb_hotwords = gr.Textbox(label="Hotwords", value="",
+                                                         info="Hotwords/hint phrases to provide the model with. Has no effect if prefix is not None.")
+                                nb_language_detection_threshold = gr.Number(label="Language Detection Threshold",
+                                                                            value=None,
+                                                                            info="If the maximum probability of the language tokens is higher than this value, the language is detected.")
+                                nb_language_detection_segments = gr.Number(label="Language Detection Segments", value=1,
+                                                                           precision=0,
+                                                                           info="Number of segments to consider for the language detection.")
                     with gr.Accordion("VAD", open=False):
                         cb_vad_filter = gr.Checkbox(label="Enable Silero VAD Filter", value=False, interactive=True)
-                        sd_threshold = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label="Speech Threshold", value=0.5, info="Lower it to be more sensitive to small sounds.")
-                        nb_min_speech_duration_ms = gr.Number(label="Minimum Speech Duration (ms)", precision=0, value=250)
+                        sd_threshold = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label="Speech Threshold",
+                                                 value=0.5, info="Lower it to be more sensitive to small sounds.")
+                        nb_min_speech_duration_ms = gr.Number(label="Minimum Speech Duration (ms)", precision=0,
+                                                              value=250)
                         nb_max_speech_duration_s = gr.Number(label="Maximum Speech Duration (s)", value=9999)
-                        nb_min_silence_duration_ms = gr.Number(label="Minimum Silence Duration (ms)", precision=0, value=2000)
+                        nb_min_silence_duration_ms = gr.Number(label="Minimum Silence Duration (ms)", precision=0,
+                                                               value=2000)
                         nb_speech_pad_ms = gr.Number(label="Speech Padding (ms)", precision=0, value=400)
                     with gr.Accordion("Diarization", open=False):
                         cb_diarize = gr.Checkbox(label="Enable Diarization")
                         tb_hf_token = gr.Text(label="HuggingFace Token", value="",
                                               info="This is only needed the first time you download the model. If you already have models, you don't need to enter. "
-                                                    "To download the model, you must manually go to \"https://huggingface.co/pyannote/speaker-diarization-3.1\" and agree to their requirement.")
-                        dd_diarization_device = gr.Dropdown(label="Device", choices=self.whisper_inf.diarizer.get_available_device(), value=self.whisper_inf.diarizer.get_device())
+                                                   "To download the model, you must manually go to \"https://huggingface.co/pyannote/speaker-diarization-3.1\" and agree to their requirement.")
+                        dd_diarization_device = gr.Dropdown(label="Device",
+                                                            choices=self.whisper_inf.diarizer.get_available_device(),
+                                                            value=self.whisper_inf.diarizer.get_device())
                     with gr.Accordion("Insanely Fast Whisper Parameters", open=False,
                                       visible=isinstance(self.whisper_inf, InsanelyFastWhisperInference)):
                         nb_chunk_length_s = gr.Number(label="Chunk Lengths (sec)", value=30, precision=0)
@@ -220,30 +339,48 @@ class App:
                         btn_openfolder = gr.Button('📂', scale=1)
 
                     params = [tb_youtubelink, dd_file_format, cb_timestamp]
-                    whisper_params = WhisperParameters(model_size=dd_model,
-                                                       lang=dd_lang,
-                                                       is_translate=cb_translate,
-                                                       beam_size=nb_beam_size,
-                                                       log_prob_threshold=nb_log_prob_threshold,
-                                                       no_speech_threshold=nb_no_speech_threshold,
-                                                       compute_type=dd_compute_type,
-                                                       best_of=nb_best_of,
-                                                       patience=nb_patience,
-                                                       condition_on_previous_text=cb_condition_on_previous_text,
-                                                       initial_prompt=tb_initial_prompt,
-                                                       temperature=sd_temperature,
-                                                       compression_ratio_threshold=nb_compression_ratio_threshold,
-                                                       vad_filter=cb_vad_filter,
-                                                       threshold=sd_threshold,
-                                                       min_speech_duration_ms=nb_min_speech_duration_ms,
-                                                       max_speech_duration_s=nb_max_speech_duration_s,
-                                                       min_silence_duration_ms=nb_min_silence_duration_ms,
-                                                       speech_pad_ms=nb_speech_pad_ms,
-                                                       chunk_length_s=nb_chunk_length_s,
-                                                       batch_size=nb_batch_size,
-                                                       is_diarize=cb_diarize,
-                                                       hf_token=tb_hf_token,
-                                                       diarization_device=dd_diarization_device)
+                    whisper_params = WhisperParameters(
+                        model_size=dd_model,
+                        lang=dd_lang,
+                        is_translate=cb_translate,
+                        beam_size=nb_beam_size,
+                        log_prob_threshold=nb_log_prob_threshold,
+                        no_speech_threshold=nb_no_speech_threshold,
+                        compute_type=dd_compute_type,
+                        best_of=nb_best_of,
+                        patience=nb_patience,
+                        condition_on_previous_text=cb_condition_on_previous_text,
+                        initial_prompt=tb_initial_prompt,
+                        temperature=sd_temperature,
+                        compression_ratio_threshold=nb_compression_ratio_threshold,
+                        vad_filter=cb_vad_filter,
+                        threshold=sd_threshold,
+                        min_speech_duration_ms=nb_min_speech_duration_ms,
+                        max_speech_duration_s=nb_max_speech_duration_s,
+                        min_silence_duration_ms=nb_min_silence_duration_ms,
+                        speech_pad_ms=nb_speech_pad_ms,
+                        chunk_length_s=nb_chunk_length_s,
+                        batch_size=nb_batch_size,
+                        is_diarize=cb_diarize,
+                        hf_token=tb_hf_token,
+                        diarization_device=dd_diarization_device,
+                        length_penalty=nb_length_penalty,
+                        repetition_penalty=nb_repetition_penalty,
+                        no_repeat_ngram_size=nb_no_repeat_ngram_size,
+                        prefix=tb_prefix,
+                        suppress_blank=cb_suppress_blank,
+                        suppress_tokens=tb_suppress_tokens,
+                        max_initial_timestamp=nb_max_initial_timestamp,
+                        word_timestamps=cb_word_timestamps,
+                        prepend_punctuations=tb_prepend_punctuations,
+                        append_punctuations=tb_append_punctuations,
+                        max_new_tokens=nb_max_new_tokens,
+                        chunk_length=nb_chunk_length,
+                        hallucination_silence_threshold=nb_hallucination_silence_threshold,
+                        hotwords=tb_hotwords,
+                        language_detection_threshold=nb_language_detection_threshold,
+                        language_detection_segments=nb_language_detection_segments
+                    )
 
                     btn_run.click(fn=self.whisper_inf.transcribe_youtube,
                                   inputs=params + whisper_params.as_list(),
@@ -266,20 +403,67 @@ class App:
                         cb_translate = gr.Checkbox(value=False, label="Translate to English?", interactive=True)
                     with gr.Accordion("Advanced Parameters", open=False):
                         nb_beam_size = gr.Number(label="Beam Size", value=1, precision=0, interactive=True)
-                        nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0, interactive=True)
+                        nb_log_prob_threshold = gr.Number(label="Log Probability Threshold", value=-1.0,
+                                                          interactive=True)
                         nb_no_speech_threshold = gr.Number(label="No Speech Threshold", value=0.6, interactive=True)
-                        dd_compute_type = gr.Dropdown(label="Compute Type", choices=self.whisper_inf.available_compute_types, value=self.whisper_inf.current_compute_type, interactive=True)
+                        dd_compute_type = gr.Dropdown(label="Compute Type",
+                                                      choices=self.whisper_inf.available_compute_types,
+                                                      value=self.whisper_inf.current_compute_type, interactive=True)
                         nb_best_of = gr.Number(label="Best Of", value=5, interactive=True)
                         nb_patience = gr.Number(label="Patience", value=1, interactive=True)
-                        cb_condition_on_previous_text = gr.Checkbox(label="Condition On Previous Text", value=True, interactive=True)
+                        cb_condition_on_previous_text = gr.Checkbox(label="Condition On Previous Text", value=True,
+                                                                    interactive=True)
                         tb_initial_prompt = gr.Textbox(label="Initial Prompt", value=None, interactive=True)
-                        sd_temperature = gr.Slider(label="Temperature", value=0, step=0.01, maximum=1.0, interactive=True)
+                        sd_temperature = gr.Slider(label="Temperature", value=0, step=0.01, maximum=1.0,
+                                                   interactive=True)
+
+                        with gr.Group(visible=isinstance(self.whisper_inf, FasterWhisperInference)):
+                            with gr.Column():
+                                nb_length_penalty = gr.Number(label="Length Penalty", value=1,
+                                                              info="Exponential length penalty constant.")
+                                nb_repetition_penalty = gr.Number(label="Repetition Penalty", value=1,
+                                                                  info="Penalty applied to the score of previously generated tokens (set > 1 to penalize).")
+                                nb_no_repeat_ngram_size = gr.Number(label="No Repeat N-gram Size", value=0, precision=0,
+                                                                    info="Prevent repetitions of n-grams with this size (set 0 to disable).")
+                                tb_prefix = gr.Textbox(label="Prefix", value="",
+                                                       info="Optional text to provide as a prefix for the first window.")
+                                cb_suppress_blank = gr.Checkbox(label="Suppress Blank", value=True,
+                                                                info="Suppress blank outputs at the beginning of the sampling.")
+                                tb_suppress_tokens = gr.Textbox(label="Suppress Tokens", value="-1",
+                                                                info="List of token IDs to suppress. -1 will suppress a default set of symbols as defined in the model config.json file.")
+                                nb_max_initial_timestamp = gr.Number(label="Max Initial Timestamp", value=1.0,
+                                                                     info="The initial timestamp cannot be later than this.")
+                                cb_word_timestamps = gr.Checkbox(label="Word Timestamps", value=False,
+                                                                 info="Extract word-level timestamps using the cross-attention pattern and dynamic time warping, and include the timestamps for each word in each segment.")
+                                tb_prepend_punctuations = gr.Textbox(label="Prepend Punctuations", value="\"'“¿([{-",
+                                                                     info="If word_timestamps is True, merge these punctuation symbols with the next word.")
+                                tb_append_punctuations = gr.Textbox(label="Append Punctuations",
+                                                                    value="\"'.。,，!！?？:：”)]}、",
+                                                                    info="If word_timestamps is True, merge these punctuation symbols with the previous word.")
+                                nb_max_new_tokens = gr.Number(label="Max New Tokens", value=None, precision=0,
+                                                              info="Maximum number of new tokens to generate per-chunk. If not set, the maximum will be set by the default max_length.")
+                                nb_chunk_length = gr.Number(label="Chunk Length", value=None,
+                                                            info="The length of audio segments. If it is not None, it will overwrite the default chunk_length of the FeatureExtractor.")
+                                nb_hallucination_silence_threshold = gr.Number(label="Hallucination Silence Threshold",
+                                                                               value=None,
+                                                                               info="When word_timestamps is True, skip silent periods longer than this threshold (in seconds) when a possible hallucination is detected.")
+                                tb_hotwords = gr.Textbox(label="Hotwords", value="",
+                                                         info="Hotwords/hint phrases to provide the model with. Has no effect if prefix is not None.")
+                                nb_language_detection_threshold = gr.Number(label="Language Detection Threshold",
+                                                                            value=None,
+                                                                            info="If the maximum probability of the language tokens is higher than this value, the language is detected.")
+                                nb_language_detection_segments = gr.Number(label="Language Detection Segments", value=1,
+                                                                           precision=0,
+                                                                           info="Number of segments to consider for the language detection.")
                     with gr.Accordion("VAD", open=False):
                         cb_vad_filter = gr.Checkbox(label="Enable Silero VAD Filter", value=False, interactive=True)
-                        sd_threshold = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label="Speech Threshold", value=0.5, info="Lower it to be more sensitive to small sounds.")
-                        nb_min_speech_duration_ms = gr.Number(label="Minimum Speech Duration (ms)", precision=0, value=250)
+                        sd_threshold = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label="Speech Threshold",
+                                                 value=0.5, info="Lower it to be more sensitive to small sounds.")
+                        nb_min_speech_duration_ms = gr.Number(label="Minimum Speech Duration (ms)", precision=0,
+                                                              value=250)
                         nb_max_speech_duration_s = gr.Number(label="Maximum Speech Duration (s)", value=9999)
-                        nb_min_silence_duration_ms = gr.Number(label="Minimum Silence Duration (ms)", precision=0, value=2000)
+                        nb_min_silence_duration_ms = gr.Number(label="Minimum Silence Duration (ms)", precision=0,
+                                                               value=2000)
                         nb_speech_pad_ms = gr.Number(label="Speech Padding (ms)", precision=0, value=400)
                     with gr.Accordion("Diarization", open=False):
                         cb_diarize = gr.Checkbox(label="Enable Diarization")
@@ -301,30 +485,48 @@ class App:
                         btn_openfolder = gr.Button('📂', scale=1)
 
                     params = [mic_input, dd_file_format]
-                    whisper_params = WhisperParameters(model_size=dd_model,
-                                                       lang=dd_lang,
-                                                       is_translate=cb_translate,
-                                                       beam_size=nb_beam_size,
-                                                       log_prob_threshold=nb_log_prob_threshold,
-                                                       no_speech_threshold=nb_no_speech_threshold,
-                                                       compute_type=dd_compute_type,
-                                                       best_of=nb_best_of,
-                                                       patience=nb_patience,
-                                                       condition_on_previous_text=cb_condition_on_previous_text,
-                                                       initial_prompt=tb_initial_prompt,
-                                                       temperature=sd_temperature,
-                                                       compression_ratio_threshold=nb_compression_ratio_threshold,
-                                                       vad_filter=cb_vad_filter,
-                                                       threshold=sd_threshold,
-                                                       min_speech_duration_ms=nb_min_speech_duration_ms,
-                                                       max_speech_duration_s=nb_max_speech_duration_s,
-                                                       min_silence_duration_ms=nb_min_silence_duration_ms,
-                                                       speech_pad_ms=nb_speech_pad_ms,
-                                                       chunk_length_s=nb_chunk_length_s,
-                                                       batch_size=nb_batch_size,
-                                                       is_diarize=cb_diarize,
-                                                       hf_token=tb_hf_token,
-                                                       diarization_device=dd_diarization_device)
+                    whisper_params = WhisperParameters(
+                        model_size=dd_model,
+                        lang=dd_lang,
+                        is_translate=cb_translate,
+                        beam_size=nb_beam_size,
+                        log_prob_threshold=nb_log_prob_threshold,
+                        no_speech_threshold=nb_no_speech_threshold,
+                        compute_type=dd_compute_type,
+                        best_of=nb_best_of,
+                        patience=nb_patience,
+                        condition_on_previous_text=cb_condition_on_previous_text,
+                        initial_prompt=tb_initial_prompt,
+                        temperature=sd_temperature,
+                        compression_ratio_threshold=nb_compression_ratio_threshold,
+                        vad_filter=cb_vad_filter,
+                        threshold=sd_threshold,
+                        min_speech_duration_ms=nb_min_speech_duration_ms,
+                        max_speech_duration_s=nb_max_speech_duration_s,
+                        min_silence_duration_ms=nb_min_silence_duration_ms,
+                        speech_pad_ms=nb_speech_pad_ms,
+                        chunk_length_s=nb_chunk_length_s,
+                        batch_size=nb_batch_size,
+                        is_diarize=cb_diarize,
+                        hf_token=tb_hf_token,
+                        diarization_device=dd_diarization_device,
+                        length_penalty=nb_length_penalty,
+                        repetition_penalty=nb_repetition_penalty,
+                        no_repeat_ngram_size=nb_no_repeat_ngram_size,
+                        prefix=tb_prefix,
+                        suppress_blank=cb_suppress_blank,
+                        suppress_tokens=tb_suppress_tokens,
+                        max_initial_timestamp=nb_max_initial_timestamp,
+                        word_timestamps=cb_word_timestamps,
+                        prepend_punctuations=tb_prepend_punctuations,
+                        append_punctuations=tb_append_punctuations,
+                        max_new_tokens=nb_max_new_tokens,
+                        chunk_length=nb_chunk_length,
+                        hallucination_silence_threshold=nb_hallucination_silence_threshold,
+                        hotwords=tb_hotwords,
+                        language_detection_threshold=nb_language_detection_threshold,
+                        language_detection_segments=nb_language_detection_segments
+                    )
 
                     btn_run.click(fn=self.whisper_inf.transcribe_mic,
                                   inputs=params + whisper_params.as_list(),
@@ -389,7 +591,8 @@ class App:
                             md_vram_table = gr.HTML(NLLB_VRAM_TABLE, elem_id="md_nllb_vram_table")
 
                     btn_run.click(fn=self.nllb_inf.translate_file,
-                                  inputs=[file_subs, dd_nllb_model, dd_nllb_sourcelang, dd_nllb_targetlang, nb_max_length, cb_timestamp],
+                                  inputs=[file_subs, dd_nllb_model, dd_nllb_sourcelang, dd_nllb_targetlang,
+                                          nb_max_length, cb_timestamp],
                                   outputs=[tb_indicator, files_subtitles])
 
                     btn_openfolder.click(fn=lambda: self.open_folder(os.path.join("outputs", "translations")),
@@ -415,7 +618,8 @@ class App:
 
 # Create the parser for command-line arguments
 parser = argparse.ArgumentParser()
-parser.add_argument('--whisper_type', type=str, default="faster-whisper", help='A type of the whisper implementation between: ["whisper", "faster-whisper", "insanely-fast-whisper"]')
+parser.add_argument('--whisper_type', type=str, default="faster-whisper",
+                    help='A type of the whisper implementation between: ["whisper", "faster-whisper", "insanely-fast-whisper"]')
 parser.add_argument('--share', type=bool, default=False, nargs='?', const=True, help='Gradio share value')
 parser.add_argument('--server_name', type=str, default=None, help='Gradio server host')
 parser.add_argument('--server_port', type=int, default=None, help='Gradio server port')
@@ -425,11 +629,17 @@ parser.add_argument('--password', type=str, default=None, help='Gradio authentic
 parser.add_argument('--theme', type=str, default=None, help='Gradio Blocks theme')
 parser.add_argument('--colab', type=bool, default=False, nargs='?', const=True, help='Is colab user or not')
 parser.add_argument('--api_open', type=bool, default=False, nargs='?', const=True, help='enable api or not')
-parser.add_argument('--whisper_model_dir', type=str, default=os.path.join("models", "Whisper"), help='Directory path of the whisper model')
-parser.add_argument('--faster_whisper_model_dir', type=str, default=os.path.join("models", "Whisper", "faster-whisper"), help='Directory path of the faster-whisper model')
-parser.add_argument('--insanely_fast_whisper_model_dir', type=str, default=os.path.join("models", "Whisper", "insanely-fast-whisper"), help='Directory path of the insanely-fast-whisper model')
-parser.add_argument('--diarization_model_dir', type=str, default=os.path.join("models", "Diarization"), help='Directory path of the diarization model')
-parser.add_argument('--nllb_model_dir', type=str, default=os.path.join("models", "NLLB"), help='Directory path of the Facebook NLLB model')
+parser.add_argument('--whisper_model_dir', type=str, default=os.path.join("models", "Whisper"),
+                    help='Directory path of the whisper model')
+parser.add_argument('--faster_whisper_model_dir', type=str, default=os.path.join("models", "Whisper", "faster-whisper"),
+                    help='Directory path of the faster-whisper model')
+parser.add_argument('--insanely_fast_whisper_model_dir', type=str,
+                    default=os.path.join("models", "Whisper", "insanely-fast-whisper"),
+                    help='Directory path of the insanely-fast-whisper model')
+parser.add_argument('--diarization_model_dir', type=str, default=os.path.join("models", "Diarization"),
+                    help='Directory path of the diarization model')
+parser.add_argument('--nllb_model_dir', type=str, default=os.path.join("models", "NLLB"),
+                    help='Directory path of the Facebook NLLB model')
 parser.add_argument('--output_dir', type=str, default=os.path.join("outputs"), help='Directory path of the outputs')
 _args = parser.parse_args()