[FEAT][faster whisperx]

requirements.txt

@@ -17,4 +17,5 @@ bitsandbytes
 peft
 accelerate
 huggingface_hub[hf_transfer]
-huggingface-hub
+huggingface-hub
+whisperx

servers/whisperx/download_mp3.py

@@ -1,21 +1,23 @@
 import os
 from pytube import YouTube
 
+
 def download_youtube_audio(video_url, output_folder):
     try:
         yt = YouTube(video_url)
         audio_stream = yt.streams.filter(only_audio=True).first()
         if audio_stream:
             output_file = audio_stream.download(output_path=output_folder)
             # Rename the file to have the .mp3 extension
-            mp3_file = output_file.split('.')[0] + '.mp3'
+            mp3_file = output_file.split(".")[0] + ".mp3"
             os.rename(output_file, mp3_file)
             print("Audio downloaded successfully:", mp3_file)
         else:
             print("No audio stream available for the given URL.")
     except Exception as e:
         print("Error:", e)
 
+
 # Example usage
 video_url = "https://www.youtube.com/watch?v=dQw4w9WgXcQ"  # Example YouTube video URL
 output_folder = "youtube_downloads"  # Output folder where the MP3 will be saved

servers/whisperx/example_api.py

@@ -15,9 +15,7 @@ class WhisperTranscription(BaseModel):
 
 
 # Construct the request data
-request_data = WhisperTranscription(
-    file = "song.mp3"
-)
+request_data = WhisperTranscription(file="song.mp3")
 
 # Specify the URL of your FastAPI application
 url = "https://localhost:8000/v1/audio/transcriptions"

servers/whisperx/whisper_model.py

@@ -0,0 +1,74 @@
+import logging
+from faster_whisper import WhisperModel
+
+
+class FasterWhisperTranscriber:
+    def __init__(
+        self,
+        model_size="large-v3",
+        device="cuda",
+        compute_type="float16",
+        model_type="faster-whisper",
+        **kwargs,
+    ):
+        """
+        Initialize the WhisperModel with specified configuration.
+
+        :param model_size: Size of the Whisper model (e.g., 'large-v3', 'distil-large-v2')
+        :param device: Computation device ('cuda' or 'cpu')
+        :param compute_type: Type of computation ('float16', 'int8_float16', 'int8')
+        :param model_type: Type of model ('faster-whisper' or 'faster-distil-whisper')
+        :param kwargs: Additional arguments for WhisperModel transcribe method
+        """
+        self.model_size = model_size
+        self.device = device
+        self.compute_type = compute_type
+        self.model_type = model_type
+        self.transcribe_options = kwargs
+        self.model = WhisperModel(
+            self.model_size, device=self.device, compute_type=self.compute_type
+        )
+
+    def run(self, task: str, *args, **kwargs):
+        """
+        Transcribes the given audio file using the Whisper model.
+
+        :param audio_file_path: Path to the audio file to be transcribed
+        :return: Transcription results
+        """
+        segments, info = self.model.transcribe(
+            task, **self.transcribe_options
+        )
+
+        # Printing language detection information
+        print(
+            f"Detected language '{info.language}' with probability {info.language_probability:.2f}"
+        )
+
+        # Handling transcription based on the model type
+        if self.model_type == "faster-whisper":
+            for segment in segments:
+                print(f"[{segment.start:.2fs} -> {segment.end:.2fs}] {segment.text}")
+        elif (
+            self.model_type == "faster-distil-whisper"
+            and "word_timestamps" in self.transcribe_options
+            and self.transcribe_options["word_timestamps"]
+        ):
+            for segment in segments:
+                for word in segment.words:
+                    print(f"[{word.start:.2fs} -> {word.end:.2fs}] {word.word}")
+        else:
+            for segment in segments:
+                print(f"[{segment.start:.2fs} -> {segment.end:.2fs}] {segment.text}")
+
+
+# Example usage
+if __name__ == "__main__":
+    logging.basicConfig()
+    logging.getLogger("faster_whisper").setLevel(logging.DEBUG)
+
+    # Example for faster-whisper with GPU and FP16
+    transcriber = FasterWhisperTranscriber(
+        model_size="large-v3", device="cuda", compute_type="float16", beam_size=5
+    )
+    transcriber.run("song.mp3")

servers/whisperx/whisperx.py

@@ -17,7 +17,6 @@
 hf_token = os.getenv("HF_TOKEN")
 
 
-
 class WhisperTranscriber:
     """
     A class for transcribing audio using the Whisper ASR system.
@@ -285,7 +284,7 @@ async def create_audio_completion(request: WhisperTranscription):
     # Log the entry into supabase
 
     transcriber = WhisperTranscriber(
-        device = "cuda" if torch.cuda.is_available() else "cpu",
+        device="cuda" if torch.cuda.is_available() else "cpu",
     )
 
     # Run the audio processing pipeline

servers/whisperx/whisperx_no_api.py

@@ -4,7 +4,7 @@
 
 import torch
 import whisperx
-import os 
+import os
 
 from dotenv import load_dotenv
 
@@ -157,8 +157,7 @@ def run(self, audio_file: str):
 
 
 # Instantiate the WhisperTranscriber
-model = WhisperTranscriber(
-)
+model = WhisperTranscriber()
 
 # Run the audio processing pipeline
 result = model.run("song.mp3")