Spaces:
Running
Running
fix bugs
Browse files
app.py
CHANGED
@@ -3,6 +3,10 @@ import queue
|
|
3 |
from huggingface_hub import snapshot_download
|
4 |
import hydra
|
5 |
import numpy as np
|
|
|
|
|
|
|
|
|
6 |
|
7 |
# Download if not exists
|
8 |
os.makedirs("checkpoints", exist_ok=True)
|
@@ -203,7 +207,7 @@ def inference_with_auto_rerank(
|
|
203 |
best_sample_rate = None
|
204 |
|
205 |
for attempt in range(max_attempts):
|
206 |
-
|
207 |
text,
|
208 |
enable_reference_audio,
|
209 |
reference_audio,
|
@@ -216,16 +220,6 @@ def inference_with_auto_rerank(
|
|
216 |
streaming=False,
|
217 |
)
|
218 |
|
219 |
-
# 获取音频数据
|
220 |
-
result = None
|
221 |
-
for item in audio_generator:
|
222 |
-
result = item
|
223 |
-
|
224 |
-
if result is None:
|
225 |
-
return None, None, "No audio generated"
|
226 |
-
|
227 |
-
_, (sample_rate, audio), message = result
|
228 |
-
|
229 |
if audio is None:
|
230 |
return None, None, message
|
231 |
|
@@ -234,6 +228,7 @@ def inference_with_auto_rerank(
|
|
234 |
|
235 |
asr_result = batch_asr(asr_model, [audio], sample_rate)[0]
|
236 |
wer = calculate_wer(text, asr_result["text"])
|
|
|
237 |
if wer <= 0.3 and not asr_result["huge_gap"]:
|
238 |
return None, (sample_rate, audio), None
|
239 |
|
@@ -253,7 +248,6 @@ n_audios = 4
|
|
253 |
global_audio_list = []
|
254 |
global_error_list = []
|
255 |
|
256 |
-
|
257 |
def inference_wrapper(
|
258 |
text,
|
259 |
enable_reference_audio,
|
|
|
3 |
from huggingface_hub import snapshot_download
|
4 |
import hydra
|
5 |
import numpy as np
|
6 |
+
import wave
|
7 |
+
import io
|
8 |
+
import pyrootutils
|
9 |
+
import gc
|
10 |
|
11 |
# Download if not exists
|
12 |
os.makedirs("checkpoints", exist_ok=True)
|
|
|
207 |
best_sample_rate = None
|
208 |
|
209 |
for attempt in range(max_attempts):
|
210 |
+
_, (sample_rate, audio), message = inference(
|
211 |
text,
|
212 |
enable_reference_audio,
|
213 |
reference_audio,
|
|
|
220 |
streaming=False,
|
221 |
)
|
222 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
223 |
if audio is None:
|
224 |
return None, None, message
|
225 |
|
|
|
228 |
|
229 |
asr_result = batch_asr(asr_model, [audio], sample_rate)[0]
|
230 |
wer = calculate_wer(text, asr_result["text"])
|
231 |
+
|
232 |
if wer <= 0.3 and not asr_result["huge_gap"]:
|
233 |
return None, (sample_rate, audio), None
|
234 |
|
|
|
248 |
global_audio_list = []
|
249 |
global_error_list = []
|
250 |
|
|
|
251 |
def inference_wrapper(
|
252 |
text,
|
253 |
enable_reference_audio,
|