please

app.py

@@ -3,38 +3,53 @@ import numpy as np
 import torch
 from datasets import load_dataset
 
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
+from transformers import pipeline, AutoTokenizer, VitsModel, VitsTokenizer
 
 
 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)
+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")
+# 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)
+# 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)
+# embeddings_dataset = load_dataset(
+#    "Matthijs/cmu-arctic-xvectors", split="validation")
+# speaker_embeddings = torch.tensor(
+# embeddings_dataset[7306]["xvector"]).unsqueeze(0)
+
+
+model_nld = VitsModel.from_pretrained("facebook/mms-tts-nld")
+tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-nld")
 
 
 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": "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)
+def synthesise2(text):
+    # inputs = processor(text=text, return_tensors="pt")
+    # speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
+    inputs = tokenizer(text, return_tensors="pt")
+    input_ids = inputs["input_ids"]
+    with torch.no_grad():
+        outputs = model_nld(input_ids)
+        speech = outputs["waveform"]
     return speech.cpu()
 
 
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
-    synthesised_speech = synthesise(translated_text)
+    synthesised_speech = synthesise2(translated_text)
     synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
     return 16000, synthesised_speech
 
@@ -44,9 +59,10 @@ 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:
 
-![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
+![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Digram of cascaded speech to speech translation")
 """
 
+
 demo = gr.Blocks()
 
 mic_translate = gr.Interface(
@@ -67,6 +83,7 @@ file_translate = gr.Interface(
 )
 
 with demo:
-    gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
+    gr.TabbedInterface([mic_translate, file_translate],
+                       ["Microphone", "Audio File"])
 
-demo.launch()
+demo.launch()

requirements.txt

@@ -1,4 +1,4 @@
 torch
-git+https://github.com/huggingface/transformers
+git+https://github.com/hollance/transformers.git@6900e8ba6532162a8613d2270ec2286c3f58f57b
 datasets
 sentencepiece