Update server.py

server.py

@@ -3,7 +3,7 @@ from io import BytesIO
 import torch
 from av import open as avopen
 from typing import Dict, List
-
+import re_matching
 import utils
 from infer import infer, get_net_g, latest_version
 from scipy.io import wavfile
@@ -67,6 +67,35 @@ for model in models:
         )
     )
 
+def generate_audio(
+    slices,
+    sdp_ratio,
+    noise_scale,
+    noise_scale_w,
+    length_scale,
+    speaker,
+    language,
+):
+    audio_list = []
+    silence = np.zeros(hps.data.sampling_rate // 2, dtype=np.int16)
+    with torch.no_grad():
+        for piece in slices:
+            audio = infer(
+                piece,
+                sdp_ratio=sdp_ratio,
+                noise_scale=noise_scale,
+                noise_scale_w=noise_scale_w,
+                length_scale=length_scale,
+                sid=speaker,
+                language=language,
+                hps=hps,
+                net_g=net_g,
+                device=device,
+            )
+            audio16bit = gr.processing_utils.convert_to_16_bit_wav(audio)
+            audio_list.append(audio16bit)
+            audio_list.append(silence)  # 将静音添加到列表中
+    return audio_list
 
 @app.route("/")
 def main():
@@ -97,23 +126,45 @@ def main():
     if speaker_id is not None:
         if speaker_id.isdigit():
             speaker = chrsMap[model][int(speaker_id)]
-
+    audio_list = []
-    with torch.no_grad():
+    if language == "mix":
-        audio = infer(
+        bool_valid, str_valid = re_matching.validate_text(text)
-            text=text,
+        if not bool_valid:
-            sdp_ratio=sdp_ratio,
+            return str_valid, (
-            noise_scale=noise,
+                hps.data.sampling_rate,
-            noise_scale_w=noisew,
+                np.concatenate([np.zeros(hps.data.sampling_rate // 2)]),
-            length_scale=length,
+            )
-            sid=speaker,
+        result = re_matching.text_matching(text)
-            language=models[model]["language"],
+        for one in result:
-            hps=hps_List[model],
+            _speaker = one.pop()
-            net_g=net_g_List[model],
+            for lang, content in one:
-            device=models[model]["device"],
+                audio_list.extend(
+                    generate_audio(
+                        content.split("|"),
+                        sdp_ratio,
+                        noise_scale,
+                        noise_scale_w,
+                        length_scale,
+                        _speaker,
+                        lang,
+                    )
+                )
+    else:
+        audio_list.extend(
+            generate_audio(
+                text.split("|"),
+                sdp_ratio,
+                noise_scale,
+                noise_scale_w,
+                length_scale,
+                speaker,
+                language,
+            )
         )
 
+    audio_concat = np.concatenate(audio_list)
     with BytesIO() as wav:
-        wavfile.write(wav, hps_List[model].data.sampling_rate, audio)
+        wavfile.write(wav, hps_List[model].data.sampling_rate, audio_concat)
         torch.cuda.empty_cache()
         if fmt == "wav":
             return Response(wav.getvalue(), mimetype="audio/wav")