Update app.py

app.py

@@ -3,33 +3,35 @@ import numpy as np
 import torch
 from datasets import load_dataset
 
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
+from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline, WhisperProcessor, BarkModel, BarkProcessor
 
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
+asr_model_id = "openai/whisper-base"
+asr_processor = WhisperProcessor.from_pretrained(asr_model_id, language="es", task="transcribe")
 # load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
+asr_pipe = pipeline(
+    "automatic-speech-recognition", 
+    model=asr_model_id, 
+    feature_extractor=asr_processor.feature_extractor,
+    tokenizer=asr_processor.tokenizer,
+    device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
-processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
-
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
-vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
-
-embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
-speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
+processor = BarkProcessor.from_pretrained("suno/bark-small")
+model = BarkModel.from_pretrained("suno/bark-small").to(device)
 
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "es", "forced_decoder_ids": asr_processor.tokenizer.get_decoder_prompt_ids(language="es", task="translate")})
     return outputs["text"]
 
 
 def synthesise(text):
-    inputs = processor(text=text, return_tensors="pt")
-    speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
-    return speech.cpu()
+    inputs = processor(text, voice_preset="v2/es_speaker_3")
+    speech = model.generate(**inputs).cpu().numpy()
+    return speech
 
 
 def speech_to_speech_translation(audio):
@@ -69,4 +71,4 @@ file_translate = gr.Interface(
 with demo:
     gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
 
-demo.launch()
+demo.launch()