Hugging Face's logo

Spaces:
mageec
/
whisper-lang-id-capstone
Runtime error

App Files Community
whisper-lang-id-capstone
File size: 2,673 Bytes 
d1e96e8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
48da169
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
 
import torch
import torch.nn.functional as F

from transformers import WhisperForConditionalGeneration, WhisperProcessor
from transformers.models.whisper.tokenization_whisper import LANGUAGES
from transformers.pipelines.audio_utils import ffmpeg_read

import gradio as gr


model_id = "mageec/whisper-tiny-hi-capstone"

device = "cuda" if torch.cuda.is_available() else "cpu"

processor = WhisperProcessor.from_pretrained(model_id)
model = WhisperForConditionalGeneration.from_pretrained(model_id)
model.eval()
model.to(device)

sampling_rate = processor.feature_extractor.sampling_rate

bos_token_id = processor.tokenizer.all_special_ids[-106]
decoder_input_ids = torch.tensor([1,bos_token_id]).to(device)


def process_audio_file(file):
    with open(file, "rb") as f:
        inputs = f.read()

    audio = ffmpeg_read(inputs, sampling_rate)
    return audio


def transcribe(Microphone, File_Upload):
    warn_output = ""
    if (Microphone is not None) and (File_Upload is not None):
        warn_output = "WARNING: You've uploaded an audio file and used the microphone. " \
                      "The recorded file from the microphone will be used and the uploaded audio will be discarded.\n"
        file = Microphone

    elif (Microphone is None) and (File_Upload is None):
        return "ERROR: You have to either use the microphone or upload an audio file"

    elif Microphone is not None:
        file = Microphone
    else:
        file = File_Upload

    audio_data = process_audio_file(file)

    input_features = processor(audio_data, return_tensors="pt").input_features
    
    with torch.no_grad():
        logits = model.forward(input_features.to(device), decoder_input_ids=decoder_input_ids).logits
    
    pred_ids = torch.argmax(logits, dim=-1)
    probability = F.softmax(logits, dim=-1).max()

    lang_ids = processor.decode(pred_ids[0])
    
    lang_ids = lang_ids.lstrip("<|").rstrip("|>")
    language = LANGUAGES.get(lang_ids, "not detected")

    return language.capitalize(), probability.cpu().numpy()


iface = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.inputs.Audio(source="microphone", type='filepath', optional=True),
        gr.inputs.Audio(source="upload", type='filepath', optional=True),
    ],
    outputs=[
        gr.outputs.Textbox(label="Language"),
        gr.Number(label="Probability"),
    ],
    layout="horizontal",
    theme="huggingface",
    title="Whisper Language Identification",
    description="Demo for Language Identification using OpenAI's [Whisper Large V2](https://huggingface.co/openai/whisper-large-v2).",
    allow_flagging='never',
)
iface.launch(enable_queue=True, server_name="ubuntu")