Update app.py

app.py

@@ -1,72 +1,76 @@
 import gradio as gr
 import numpy as np
 import torch
-from datasets import load_dataset
-
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
-
-
+from transformers import AutoProcessor, pipeline, BarkModel
+
+ASR_MODEL_NAME = "bofenghuang/whisper-large-v2-cv11-german"
+TTS_MODEL_NAME = "suno/bark-small"
+BATCH_SIZE = 8
+voices = {
+"male" : "v2/en_speaker_6",
+"female" : "v2/en_speaker_9"
+}
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
 # load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", 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)
+asr_pipe = pipeline("automatic-speech-recognition", model=ASR_MODEL_NAME, chunk_length_s=10,device=device)
 
+# load text-to-speech checkpoint
+processor = AutoProcessor.from_pretrained("suno/bark-small")
+model = BarkModel.from_pretrained("suno/bark-small").to(device)
+sampling_rate = model.generation_config.sample_rate
 
 def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"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()
-
-
-def speech_to_speech_translation(audio):
+    
+def synthesise(text, voice_preset):
+    inputs = processor(text=text, return_tensors="pt",voice_preset=voice_preset)
+    speech = model.generate(**inputs.to(device))
+    return speech[0]
+    
+def speech_to_speech_translation(audio, voice):
+    voice_preset = None
     translated_text = translate(audio)
-    synthesised_speech = synthesise(translated_text)
-    synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
-    return 16000, synthesised_speech
-
-
-title = "Cascaded STST"
+    print(translated_text)
+    if voice == "Female":
+     voice_preset = voices["female"]
+    else:
+     voice_preset = voices["male"]
+    synthesised_speech = synthesise(translated_text, voice_preset)
+    synthesised_speech = (synthesised_speech.cpu().numpy() * 32767).astype(np.int16)
+    return sampling_rate, synthesised_speech
+    
+title = "Cascaded STST - Any language to German speech"
 description = """
-Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in English. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's
-[SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model for text-to-speech:
-
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in German. Demo uses fine-tuned version of openai/whisper-large-v2 model (https://huggingface.co/bofenghuang/whisper-large-v2-cv11-german) for speech translation, and Suno's
+[Bark-large](https://huggingface.co/suno/bark-small) model for text-to-speech:
 ![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
 """
-
 demo = gr.Blocks()
 
 mic_translate = gr.Interface(
     fn=speech_to_speech_translation,
-    inputs=gr.Audio(source="microphone", type="filepath"),
+    inputs=[gr.Audio(source="microphone", type="filepath"),
+    gr.inputs.Radio(["Male", "Female"], label="Voice", default="Male")],
     outputs=gr.Audio(label="Generated Speech", type="numpy"),
     title=title,
     description=description,
+    allow_flagging="never"
 )
 
 file_translate = gr.Interface(
     fn=speech_to_speech_translation,
-    inputs=gr.Audio(source="upload", type="filepath"),
+    inputs=[gr.Audio(source="upload", type="filepath"),
+    gr.inputs.Radio(["Male", "Female"], label="Voice", default="Male")],
     outputs=gr.Audio(label="Generated Speech", type="numpy"),
-    examples=[["./example.wav"]],
     title=title,
     description=description,
+    allow_flagging="never"
 )
 
 with demo:
     gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
 
-demo.launch()
+demo.queue(concurrency_count=2,max_size=10)
+demo.launch()