Initial commit: Audio transcription app with Qwen-Omni-7B and speaker diarization

requirements.txt

+torch
+transformers
+pyannote.audio
+librosa
+bitsandbytes
+accelerate
\ No newline at end of file

transcript.py

+import argparse
+import torch
+from transformers import AutoModelForCausalLM, AutoProcessor, BitsAndBytesConfig
+from pyannote.audio import Pipeline
+import librosa
+import os
+
+def main():
+    parser = argparse.ArgumentParser(description='Transcribe audio with speakers and timestamps using Qwen-Omni-7B')
+    parser.add_argument('audio_file', help='Path to the audio file')
+    args = parser.parse_args()
+
+    audio_file = args.audio_file
+
+    # Check if file exists
+    if not os.path.exists(audio_file):
+        print(f"Error: Audio file '{audio_file}' not found.")
+        return
+
+    # Load Qwen-Omni-7B model with 4-bit quantization
+    quantization_config = BitsAndBytesConfig(load_in_4bit=True)
+    model = AutoModelForCausalLM.from_pretrained(
+        "Qwen/Qwen-Omni-7B",
+        quantization_config=quantization_config,
+        device_map="auto"
+    )
+    processor = AutoProcessor.from_pretrained("Qwen/Qwen-Omni-7B")
+
+    # Load diarization pipeline
+    # Note: Requires Hugging Face token set as HF_TOKEN environment variable
+    hf_token = os.getenv("HF_TOKEN")
+    if not hf_token:
+        print("Error: Please set HF_TOKEN environment variable for pyannote authentication.")
+        return
+    diarization_pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization", use_auth_token=hf_token)
+
+    # Load audio
+    audio, sr = librosa.load(audio_file, sr=16000)
+
+    # Diarize first
+    diarization = diarization_pipeline(audio_file)
+
+    # Transcribe each speaker segment
+    output_lines = []
+    for turn, _, speaker_id in diarization.itertracks(yield_label=True):
+        start = turn.start
+        end = turn.end
+
+        # Extract audio chunk
+        start_sample = int(start * sr)
+        end_sample = int(end * sr)
+        audio_chunk = audio[start_sample:end_sample]
+
+        # Prepare inputs for Qwen-Omni
+        conversation = [
+            {"role": "user", "content": [
+                {"type": "audio", "audio": {"waveform": audio_chunk, "sample_rate": sr}},
+                {"type": "text", "text": "Transcribe this audio segment exactly as spoken."}
+            ]}
+        ]
+        inputs = processor(conversation=conversation, return_tensors="pt").to(model.device)
+
+        # Generate transcription
+        with torch.no_grad():
+            generated_ids = model.generate(**inputs, max_new_tokens=200, do_sample=False)
+        text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
+
+        # Format timestamps
+        start_min, start_sec = divmod(start, 60)
+        start_hour, start_min = divmod(start_min, 60)
+        start_str = f"{int(start_hour):02d}:{int(start_min):02d}:{start_sec:05.2f}"
+
+        end_min, end_sec = divmod(end, 60)
+        end_hour, end_min = divmod(end_min, 60)
+        end_str = f"{int(end_hour):02d}:{int(end_min):02d}:{end_sec:05.2f}"
+
+        line = f"[{start_str} - {end_str}] {speaker_id}: {text}"
+        output_lines.append(line)
+
+    # Write to output file
+    base_name = os.path.splitext(audio_file)[0]
+    output_file = base_name + '.txt'
+    with open(output_file, 'w', encoding='utf-8') as f:
+        f.write('\n'.join(output_lines))
+
+    print(f"Transcription saved to {output_file}")
+
+if __name__ == "__main__":
+    main()
\ No newline at end of file