EngineX-Iluvatar/enginex-mr_series-asr
15
1
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Fix model output Unicode U+2581 separator related issues

Browse Source
This commit is contained in:
sunjichen
2026-02-11 16:39:16 +08:00
parent 4aebd0a465
commit a1609fb499
1 changed files with 5 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
fastapi_funasr.py
View File

@@ -168,6 +168,7 @@ def test_funasr(audio_file, lang):
segment_path = f"temp_seg_{i}.wav" segment_path = f"temp_seg_{i}.wav"
torchaudio.save(segment_path, segment, sample_rate) torchaudio.save(segment_path, segment, sample_rate)
ts1 = time.time() ts1 = time.time()
text = None
if model_type == "sensevoice": if model_type == "sensevoice":
res = model.generate( res = model.generate(
input=segment_path, input=segment_path,
@@ -203,14 +204,12 @@ def test_funasr(audio_file, lang):
# paraformer模型会一个字一个字输出，中间夹太多空格会影响1-cer的结果 # paraformer模型会一个字一个字输出，中间夹太多空格会影响1-cer的结果
if lang == "zh": if lang == "zh":
text = text.replace(" ", "") text = text.replace(" ", "")
text = text.replace("_", " ")
elif model_type == "conformer": elif model_type == "conformer":
res = model.generate( res = model.generate(
input=segment_path, input=segment_path,
batch_size_s=300 batch_size_s=300
) )
text = res[0]["text"] text = res[0]["text"]
text = text.replace("_", " ")
# elif model_type == "uni_asr": # elif model_type == "uni_asr":
# if i == 0: # if i == 0:
# os.remove(segment_path) # os.remove(segment_path)
@@ -221,6 +220,10 @@ def test_funasr(audio_file, lang):
# text = res[0]["text"] # text = res[0]["text"]
else: else:
raise RuntimeError("unknown model type") raise RuntimeError("unknown model type")
if text is not None:
# some models output "▁" (9601, Unicode U+2581) as separator between words, replace them with space for better readability
text = text.replace("_", " ")
text = text.replace(chr(9601), " ")
ts2 = time.time() ts2 = time.time()
generated_text += text generated_text += text
processing_time += (ts2 - ts1) processing_time += (ts2 - ts1)