Spaces:
Runtime error
Runtime error
Rename temp_store to temp_store.py
Browse files- temp_store +0 -0
- temp_store.py +60 -0
temp_store
DELETED
File without changes
|
temp_store.py
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
import gradio as gr
|
3 |
+
import torch
|
4 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, VitsModel
|
5 |
+
from nemo.collections.asr.models import EncDecMultiTaskModel
|
6 |
+
|
7 |
+
|
8 |
+
# load speech to text model
|
9 |
+
canary_model = EncDecMultiTaskModel.from_pretrained('nvidia/canary-1b')
|
10 |
+
canary_model.eval()
|
11 |
+
canary_model.to('cpu')
|
12 |
+
|
13 |
+
# update decode params
|
14 |
+
canary_model.change_decoding_strategy(None)
|
15 |
+
decode_cfg = canary_model.cfg.decoding
|
16 |
+
decode_cfg.beam.beam_size = 1
|
17 |
+
canary_model.change_decoding_strategy(decode_cfg)
|
18 |
+
|
19 |
+
|
20 |
+
|
21 |
+
|
22 |
+
# Load the text processing model and tokenizer
|
23 |
+
proc_tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct")
|
24 |
+
proc_model = AutoModelForCausalLM.from_pretrained(
|
25 |
+
"microsoft/Phi-3-mini-4k-instruct",
|
26 |
+
trust_remote_code=True,
|
27 |
+
)
|
28 |
+
proc_model.eval()
|
29 |
+
proc_model.to('cpu')
|
30 |
+
|
31 |
+
|
32 |
+
# Load the TTS model
|
33 |
+
tts_model = VitsModel.from_pretrained("facebook/mms-tts-eng")
|
34 |
+
tts_tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")
|
35 |
+
tts_model.eval()
|
36 |
+
tts_model.to('cpu')
|
37 |
+
|
38 |
+
|
39 |
+
def process_speech(speech):
|
40 |
+
# Convert the speech to text
|
41 |
+
transcription = canary_model.transcribe(
|
42 |
+
speech,
|
43 |
+
logprobs=False,
|
44 |
+
)
|
45 |
+
|
46 |
+
# Process the text
|
47 |
+
inputs = proc_tokenizer.encode(transcription + proc_tokenizer.eos_token, return_tensors='pt')
|
48 |
+
outputs = proc_model.generate(inputs, max_length=100, temperature=0.7, pad_token_id=proc_tokenizer.eos_token_id)
|
49 |
+
text = proc_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
50 |
+
processed_text = tts_tokenizer(text, return_tensors="pt")
|
51 |
+
|
52 |
+
# Convert the processed text to speech
|
53 |
+
with torch.no_grad():
|
54 |
+
audio = tts_model(**inputs).waveform
|
55 |
+
|
56 |
+
return audio
|
57 |
+
|
58 |
+
iface = gr.Interface(fn=process_speech, inputs=gr.inputs.Audio(source="microphone"), outputs="audio")
|
59 |
+
|
60 |
+
iface.launch()
|