Spaces:
Runtime error
Runtime error
Fix generate_kwargs and sourced model
Browse files
app.py
CHANGED
@@ -9,7 +9,7 @@ from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Proce
|
|
9 |
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
10 |
|
11 |
# load speech translation checkpoint
|
12 |
-
asr_pipe = pipeline("automatic-speech-recognition", model="
|
13 |
|
14 |
# load text-to-speech checkpoint and speaker embeddings
|
15 |
processor = SpeechT5Processor.from_pretrained("WasuratS/speecht5_finetuned_voxpopuli_nl")
|
@@ -22,7 +22,7 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
|
|
22 |
|
23 |
|
24 |
def translate(audio):
|
25 |
-
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "
|
26 |
return outputs["text"]
|
27 |
|
28 |
|
@@ -41,7 +41,7 @@ def speech_to_speech_translation(audio):
|
|
41 |
|
42 |
title = "Cascaded STST - Danish to Dutch"
|
43 |
description = """
|
44 |
-
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in Danish language to target speech in Dutch ! <br/> Demo uses
|
45 |
[SpeechT5 TTS](https://huggingface.co/WasuratS/speecht5_finetuned_voxpopuli_nl) model for text-to-speech.
|
46 |
|
47 |
![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
|
|
|
9 |
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
10 |
|
11 |
# load speech translation checkpoint
|
12 |
+
asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
|
13 |
|
14 |
# load text-to-speech checkpoint and speaker embeddings
|
15 |
processor = SpeechT5Processor.from_pretrained("WasuratS/speecht5_finetuned_voxpopuli_nl")
|
|
|
22 |
|
23 |
|
24 |
def translate(audio):
|
25 |
+
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
|
26 |
return outputs["text"]
|
27 |
|
28 |
|
|
|
41 |
|
42 |
title = "Cascaded STST - Danish to Dutch"
|
43 |
description = """
|
44 |
+
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in Danish language to target speech in Dutch ! <br/> Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and my fine tuned Microsoft's
|
45 |
[SpeechT5 TTS](https://huggingface.co/WasuratS/speecht5_finetuned_voxpopuli_nl) model for text-to-speech.
|
46 |
|
47 |
![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
|