add deep-translator

app.py

@@ -2,48 +2,54 @@ import gradio as gr
 import numpy as np
 import torch
 from datasets import load_dataset
-
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
-
+from deep_translator import GoogleTranslator
+from transformers import (
+    AutoTokenizer,
+    SpeechT5ForTextToSpeech,
+    SpeechT5HifiGan,
+    SpeechT5Processor,
+    VitsModel,
+    pipeline,
+)
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
+# device = "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")
-
-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)
+# load text-to-speech mms-tts-id model (speaker embeddings included)
+model = VitsModel.from_pretrained("facebook/mms-tts-ind")
+tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-ind")
 
 
 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)
-    return speech.cpu()
+    inputs = tokenizer(text=text, return_tensors="pt")
+    with torch.no_grad():
+        speech = model(**inputs).waveform
+    return speech.reshape(-1, 1).cpu()
 
 
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
-    synthesised_speech = synthesise(translated_text)
+    google_translated = GoogleTranslator(
+        source="en", target="id").translate(translated_text)
+    synthesised_speech = synthesise(google_translated)
     synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
     return 16000, synthesised_speech
 
 
 title = "Cascaded STST"
 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 Indonesian. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's
+[MMS TTS IND](https://huggingface.co/facebook/mms-tts-ind) 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")
 """
 
@@ -51,7 +57,7 @@ demo = gr.Blocks()
 
 mic_translate = gr.Interface(
     fn=speech_to_speech_translation,
-    inputs=gr.Audio(source="microphone", type="filepath"),
+    inputs=gr.Audio(sources="microphone", type="filepath"),
     outputs=gr.Audio(label="Generated Speech", type="numpy"),
     title=title,
     description=description,
@@ -59,7 +65,7 @@ mic_translate = gr.Interface(
 
 file_translate = gr.Interface(
     fn=speech_to_speech_translation,
-    inputs=gr.Audio(source="upload", type="filepath"),
+    inputs=gr.Audio(sources="upload", type="filepath"),
     outputs=gr.Audio(label="Generated Speech", type="numpy"),
     examples=[["./example.wav"]],
     title=title,
@@ -67,6 +73,7 @@ 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()