add simplified to traditional chinese conversion

app/asr_worker.py

@@ -1,13 +1,13 @@
 import numpy as np
+import pysoxr
+from opencc import OpenCC
 import sherpa_onnx
 from pathlib import Path
-import scipy.signal
-
-def resample_audio(audio, orig_sr, target_sr):
-    return scipy.signal.resample_poly(audio, target_sr, orig_sr)
 
 MODEL_DIR = Path("models/zipformer_bilingual")
 
+converter = OpenCC('s2t')  # Simplified to Traditional Chinese
+
 def create_recognizer():
     return sherpa_onnx.OnlineRecognizer.from_transducer(
         tokens=str(MODEL_DIR / "tokens.txt"),
@@ -21,6 +21,9 @@ def create_recognizer():
         decoding_method="greedy_search"
     )
 
+def resample_audio(audio, orig_sr, target_sr):
+    return pysoxr.resample(audio, orig_sr, target_sr)
+
 def stream_audio(raw_pcm_bytes, stream, recognizer):
     audio = np.frombuffer(raw_pcm_bytes, dtype=np.float32)
     if audio.size == 0:
@@ -33,7 +36,7 @@ def stream_audio(raw_pcm_bytes, stream, recognizer):
     if recognizer.is_ready(stream):
         recognizer.decode_streams([stream])
     result = recognizer.get_result(stream)
-    return result, rms
+    return converter.convert(result), rms
 
 def finalize_stream(stream, recognizer):
     tail = np.zeros(int(0.66 * 16000), dtype=np.float32)
@@ -41,4 +44,5 @@ def finalize_stream(stream, recognizer):
     stream.input_finished()
     while recognizer.is_ready(stream):
         recognizer.decode_streams([stream])
-    return recognizer.get_result(stream)
+    result = recognizer.get_result(stream)
+    return converter.convert(result)

requirements.txt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c564834ceb94d0d9aceef1ba37eed4451bca6eb05158ebd81518a7e68a1c2ef
-size 60
+oid sha256:4b08ab37381178cb19b7876c22c7df8a82205f03c506c487a6f0cbe773707c23
+size 88