Update app.py

app.py

@@ -17,11 +17,11 @@ YT_LENGTH_LIMIT_S = 3600  # limit to 1 hour YouTube files
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
-model_id = "Kushtrim/whisper-large-v3-turbo-shqip-115h"
+model_id = "Kushtrim/whisper-large-v3-turbo-shqip"
 model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype, use_safetensors=True).to(device)
 processor = AutoProcessor.from_pretrained(model_id)
 pipe = pipeline("automatic-speech-recognition", model=model, tokenizer=processor.tokenizer, feature_extractor=processor.feature_extractor,
-                max_new_tokens=256, chunk_length_s=15, batch_size=16, torch_dtype=torch_dtype, device=device,
+                max_new_tokens=256, chunk_length_s=28, batch_size=16, torch_dtype=torch_dtype, device=device,
                 token=os.environ["HF"])
 
 @spaces.GPU
@@ -31,7 +31,7 @@ def transcribe(inputs, task):
             "No audio file submitted! Please upload or record an audio file before submitting your request.")
 
     text = pipe(inputs, generate_kwargs={
-                "task": task, 'language': 'sq'}, return_timestamps=True)["text"]
+                'num_beams': 5, "task": task, 'language': 'sq'}, return_timestamps=True)["text"]
     return text
 
 
@@ -109,7 +109,7 @@ file_transcribe = gr.Interface(
     title="Whisper Large V3 Turbo Shqip: Transcribe Audio",
     description=("This fine-tuned Whisper model provides reliable transcription for Albanian audio, whether from a microphone or an uploaded file. " 
                  "Key details about this project:"
-                 "\n\n- Fine-tuned on 115 hours of carefully curated Albanian audio data. "
+                 "\n\n- Fine-tuned on 200 hours of carefully curated Albanian audio data. "
                  "\n- This is the third training run, reflecting continuous improvements as the dataset evolves. "
                  f"\n- Hosted on Hugging Face. Repository: [{model_id}](https://huggingface.co/{model_id}). "
                 ),
@@ -125,7 +125,7 @@ mf_transcribe = gr.Interface(
     title="Whisper Large V3 Turbo Shqip: Transcribe Audio",
     description=("This fine-tuned Whisper model provides reliable transcription for Albanian audio, whether from a microphone or an uploaded file. " 
                  "Key details about this project:"
-                 "\n\n- Fine-tuned on 115 hours of carefully curated Albanian audio data. "
+                 "\n\n- Fine-tuned on 200 hours of carefully curated Albanian audio data. "
                  "\n- This is the third training run, reflecting continuous improvements as the dataset evolves. "
                  f"\n- Hosted on Hugging Face. Repository: [{model_id}](https://huggingface.co/{model_id}). "
                 ),
@@ -142,7 +142,7 @@ yt_transcribe = gr.Interface(
     title="Whisper Large V3 Turbo Shqip: Transcribe Audio",
     description=("This fine-tuned Whisper model provides reliable transcription for Albanian audio, whether from a microphone or an uploaded file. " 
                  "Key details about this project:"
-                 "\n\n- Fine-tuned on 115 hours of carefully curated Albanian audio data. "
+                 "\n\n- Fine-tuned on 200 hours of carefully curated Albanian audio data. "
                  "\n- This is the third training run, reflecting continuous improvements as the dataset evolves. "
                  f"\n- Hosted on Hugging Face. Repository: [{model_id}](https://huggingface.co/{model_id}). "
                 ),