Add paddlespeech asr.

app.py

@@ -0,0 +1,42 @@
+import gradio as gr
+import librosa
+import numpy as np
+import paddlehub as hub
+from paddlenlp import Taskflow
+from paddlespeech.cli import ASRExecutor
+import soundfile as sf
+
+# asr_model = hub.Module(name='u2_conformer_aishell')
+asr_executor = ASRExecutor()
+text_correct_model = Taskflow("text_correction")
+punc_model = hub.Module(name='ernie_punc')
+
+
+def speech_recognize(file):
+    data, sr = librosa.load(file)
+    if sr != 16000:
+        data = librosa.resample(data, sr, 16000)
+    sf.write(file, data, samplerate=16000)
+
+    print(f'[Audio Input] shape: {data.shape}, dtype: {data.dtype}, file: {file}')
+    # text = asr_model.speech_recognize(file, device='cpu')
+    text = asr_executor('wenetspeech', 'zh', 16000, None, None, file, 'cpu')
+    text_correction = text_correct_model(text)[0]
+    cor_text, errors = text_correction['target'], text_correction['errors']
+    print(f'[Text Correction] errors: {errors}')
+    punc_text = punc_model.add_puncs(cor_text, device='cpu')[0]
+
+    ret = ''
+    ret += f'[ASR] {text}\n'
+    ret += f'[COR] {cor_text}\n'
+    ret += f'[PUN] {punc_text}'
+    return ret
+
+
+iface = gr.Interface(
+    fn=speech_recognize, 
+    inputs=gr.inputs.Audio(source="microphone", type='filepath'),
+    outputs="text",
+)
+iface.launch()
+

packages.txt

@@ -0,0 +1,2 @@
+libsndfile1
+sox

requirements.txt

@@ -0,0 +1,8 @@
+paddlepaddle
+paddlespeech
+paddlehub
+paddlenlp
+pypinyin
+SoundFile==0.9.0.post1
+librosa
+opencv-python-headless