Marco-Cheung
commited on
Commit
•
de411e0
1
Parent(s):
e9f9b4b
Update app.py
Browse files
app.py
CHANGED
@@ -15,15 +15,14 @@ device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
|
15 |
|
16 |
# load speech translation checkpoint
|
17 |
asr_pipe = pipeline("automatic-speech-recognition", model=ASR_MODEL_NAME, chunk_length_s=10,device=device)
|
|
|
|
|
18 |
|
19 |
# load text-to-speech checkpoint
|
20 |
processor = AutoProcessor.from_pretrained("suno/bark-small")
|
21 |
model = BarkModel.from_pretrained("suno/bark-small").to(device)
|
22 |
sampling_rate = model.generation_config.sample_rate
|
23 |
|
24 |
-
# set the forced ids
|
25 |
-
model.generation_config.forced_decoder_ids = processor.tokenizer.get_decoder_prompt_ids(language='de', task="translate")
|
26 |
-
|
27 |
def translate(audio):
|
28 |
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
|
29 |
return outputs["text"]
|
|
|
15 |
|
16 |
# load speech translation checkpoint
|
17 |
asr_pipe = pipeline("automatic-speech-recognition", model=ASR_MODEL_NAME, chunk_length_s=10,device=device)
|
18 |
+
# set the forced ids
|
19 |
+
asr_pipe.model.config.forced_decoder_ids = asr_pipe.tokenizer.get_decoder_prompt_ids(language='de', task="translate")
|
20 |
|
21 |
# load text-to-speech checkpoint
|
22 |
processor = AutoProcessor.from_pretrained("suno/bark-small")
|
23 |
model = BarkModel.from_pretrained("suno/bark-small").to(device)
|
24 |
sampling_rate = model.generation_config.sample_rate
|
25 |
|
|
|
|
|
|
|
26 |
def translate(audio):
|
27 |
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
|
28 |
return outputs["text"]
|