Spaces:
Sleeping
Sleeping
File size: 6,034 Bytes
d7ba1b9 c0561a7 d7ba1b9 c0561a7 d7ba1b9 b68a112 d7ba1b9 c0561a7 d7ba1b9 b68a112 0e4db4a d7ba1b9 e21f7ff d7ba1b9 0e4db4a d7ba1b9 0e4db4a d7ba1b9 e21f7ff d7ba1b9 |
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 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 |
from __future__ import annotations
import os
import gradio as gr
import torch
import torchaudio
import spaces
import nemo.collections.asr as nemo_asr
LANGUAGE_NAME_TO_CODE = {
"Hindi": "hi",
}
DESCRIPTION = """\
### **IndicConformer: Speech Recognition for Indian Languages** 🎙️➡️📜
**IndicConformer**, a speech recognition model for **22 Indian languages**. The model operates in two modes: **CTC (Connectionist Temporal Classification)** and **RNNT (Recurrent Neural Network Transducer)**
#### **How to Use:**
1. **Upload or record** an audio clip in Hindi.
2. Select the **mode** (CTC or RNNT) for transcription.
3. Click **"Transcribe"** to generate the corresponding text.
"""
hf_token = os.getenv("HF_TOKEN")
device = "cuda:0" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
torch_dtype = torch.bfloat16 if device != "cpu" else torch.float32
model_name_or_path = "ai4bharat/IndicConformer"
model = nemo_asr.models.EncDecCTCModel.from_pretrained(model_name_or_path).to(device)
# model = nemo_asr.models.EncDecCTCModel.restore_from("indicconformer_stt_bn_hybrid_rnnt_large.nemo").to(device)
model.eval()
CACHE_EXAMPLES = os.getenv("CACHE_EXAMPLES") == "1" and torch.cuda.is_available()
AUDIO_SAMPLE_RATE = 16000
MAX_INPUT_AUDIO_LENGTH = 600 # in seconds
DEFAULT_TARGET_LANGUAGE = "Hindi"
@spaces.GPU
def run_asr_ctc(input_audio: str, target_language: str) -> str:
lang_id = LANGUAGE_NAME_TO_CODE[target_language]
# Load and preprocess audio
audio_tensor, orig_freq = torchaudio.load(input_audio)
# Convert to mono if not already
if audio_tensor.shape[0] > 1:
audio_tensor = torch.mean(audio_tensor, dim=0, keepdim=True)
# Ensure shape [B x T]
if len(audio_tensor.shape) == 1:
audio_tensor = audio_tensor.unsqueeze(0) # Add batch dimension if missing
if audio_tensor.ndim > 1:
audio_tensor = audio_tensor.squeeze(0)
# Resample to 16kHz
audio_tensor = torchaudio.functional.resample(audio_tensor, orig_freq=orig_freq, new_freq=16000)
model.cur_decoder = "ctc"
ctc_text = model.transcribe([audio_tensor.numpy()], batch_size=1, logprobs=False, language_id=lang_id)[0]
return ctc_text[0]
# @spaces.GPU
# def run_asr_ctc(input_audio: str, target_language: str) -> str:
# # preprocess_audio(input_audio)
# # input_audio, orig_freq = torchaudio.load(input_audio)
# # input_audio = torchaudio.functional.resample(input_audio, orig_freq=orig_freq, new_freq=16000)
# lang_id = LANGUAGE_NAME_TO_CODE[target_language]
# model.cur_decoder = "ctc"
# ctc_text = model.transcribe([input_audio], batch_size=1, logprobs=False, language_id=lang_id)[0]
# return ctc_text[0]
@spaces.GPU
def run_asr_rnnt(input_audio: str, target_language: str) -> str:
lang_id = LANGUAGE_NAME_TO_CODE[target_language]
# Load and preprocess audio
audio_tensor, orig_freq = torchaudio.load(input_audio)
# Convert to mono if not already
if audio_tensor.shape[0] > 1:
audio_tensor = torch.mean(audio_tensor, dim=0, keepdim=True)
# Ensure shape [B x T]
if len(audio_tensor.shape) == 1:
audio_tensor = audio_tensor.unsqueeze(0) # Add batch dimension if missing
if audio_tensor.ndim > 1:
audio_tensor = audio_tensor.squeeze(0)
# Resample to 16kHz
audio_tensor = torchaudio.functional.resample(audio_tensor, orig_freq=orig_freq, new_freq=16000)
model.cur_decoder = "rnnt"
ctc_text = model.transcribe([audio_tensor.numpy()], batch_size=1, logprobs=False, language_id=lang_id)[0]
return ctc_text[0]
# @spaces.GPU
# def run_asr_rnnt(input_audio: str, target_language: str) -> str:
# # preprocess_audio(input_audio)
# # input_audio, orig_freq = torchaudio.load(input_audio)
# # input_audio = torchaudio.functional.resample(input_audio, orig_freq=orig_freq, new_freq=16000)
# lang_id = LANGUAGE_NAME_TO_CODE[target_language]
# model.cur_decoder = "rnnt"
# ctc_text = model.transcribe([input_audio], batch_size=1,logprobs=False, language_id=lang_id)[0]
# return ctc_text[0]
with gr.Blocks() as demo_asr_ctc:
with gr.Row():
with gr.Column():
with gr.Group():
input_audio = gr.Audio(label="Input speech", type="filepath")
target_language = gr.Dropdown(
label="Target language",
choices=LANGUAGE_NAME_TO_CODE.keys(),
value=DEFAULT_TARGET_LANGUAGE,
)
btn = gr.Button("Transcribe")
with gr.Column():
output_text = gr.Textbox(label="Transcribed text")
btn.click(
fn=run_asr_ctc,
inputs=[input_audio, target_language],
outputs=output_text,
api_name="asr",
)
with gr.Blocks() as demo_asr_rnnt:
with gr.Row():
with gr.Column():
with gr.Group():
input_audio = gr.Audio(label="Input speech", type="filepath")
target_language = gr.Dropdown(
label="Target language",
choices=LANGUAGE_NAME_TO_CODE.keys(),
value=DEFAULT_TARGET_LANGUAGE,
)
btn = gr.Button("Transcribe")
with gr.Column():
output_text = gr.Textbox(label="Transcribed text")
btn.click(
fn=run_asr_rnnt,
inputs=[input_audio, target_language],
outputs=output_text,
api_name="asr",
)
with gr.Blocks(css="style.css") as demo:
gr.Markdown(DESCRIPTION)
# gr.DuplicateButton(
# value="Duplicate Space for private use",
# elem_id="duplicate-button",
# visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
# )
with gr.Tabs():
with gr.Tab(label="CTC"):
demo_asr_ctc.render()
with gr.Tab(label="RNNT"):
demo_asr_rnnt.render()
if __name__ == "__main__":
demo.queue(max_size=50).launch()
|