Use mms for german

app.py

@@ -1,34 +1,47 @@
 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 (
+    VitsModel,
+    VitsTokenizer,
+    pipeline
+)
 
 
 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)
+asr_pipe = pipeline(
+    "automatic-speech-recognition",
+    model="openai/whisper-base",
+    device=device
+)
 
-embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
-speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
+model = VitsModel.from_pretrained("Matthijs/mms-tts-deu")
+tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-deu")
 
 
 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": "de"}
+    )
     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)
+    if len(text.strip()) == 0:
+        return (16000, np.zeros(0).astype(np.int16))
+
+    inputs = tokenizer(text, return_tensors="pt")
+    input_ids = inputs["input_ids"]
+
+    with torch.no_grad():
+        outputs = model(input_ids)
+
+    speech = outputs.audio[0]
     return speech.cpu()
 
 
@@ -56,7 +69,6 @@ mic_translate = gr.Interface(
     title=title,
     description=description,
 )
-
 file_translate = gr.Interface(
     fn=speech_to_speech_translation,
     inputs=gr.Audio(source="upload", type="filepath"),
@@ -70,3 +82,4 @@ with demo:
     gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
 
 demo.launch()
+