first update

.gitignore

@@ -0,0 +1,5 @@
+/__pycache__
+/flagged
+
+.env
+/testing

app.py

@@ -9,26 +9,33 @@ from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Proce
 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")
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
-vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+model = SpeechT5ForTextToSpeech.from_pretrained(
+    "sanchit-gandhi/speecht5_tts_vox_nl").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)
 
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
+    outputs = asr_pipe(audio, max_new_tokens=128, generate_kwargs={
+                       "task": "transcribe", "language": "nl"})
     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)
+    speech = model.generate_speech(inputs["input_ids"].to(
+        device), speaker_embeddings.to(device), vocoder=vocoder)
     return speech.cpu()
 
 
@@ -41,7 +48,7 @@ def speech_to_speech_translation(audio):
 
 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
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in Dutch. 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")
@@ -49,14 +56,6 @@ Demo for cascaded speech-to-speech translation (STST), mapping from source speec
 
 demo = gr.Blocks()
 
-mic_translate = gr.Interface(
-    fn=speech_to_speech_translation,
-    inputs=gr.Audio(source="microphone", type="filepath"),
-    outputs=gr.Audio(label="Generated Speech", type="numpy"),
-    title=title,
-    description=description,
-)
-
 file_translate = gr.Interface(
     fn=speech_to_speech_translation,
     inputs=gr.Audio(source="upload", type="filepath"),
@@ -67,6 +66,7 @@ file_translate = gr.Interface(
 )
 
 with demo:
-    gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
+    gr.TabbedInterface([file_translate],
+                       ["Audio File"])
 
 demo.launch()