rixty/KoboldAI-Client
1
1
Fork 0
mirror of https://github.com/KoboldAI/KoboldAI-Client.git synced 2025-06-05 21:59:24 +02:00
Code Issues Projects Releases Wiki Activity

Merge branch 'Model_Plugins' of https://github.com/ebolam/KoboldAI into Model_Plugins

Browse Source
This commit is contained in:
ebolam
2023-09-27 13:58:15 -04:00
parent 8944f69e4b b9600e4329
commit 67af0ce102
1 changed files with 15 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
koboldai_settings.py
View File

@@ -2116,6 +2116,7 @@ class KoboldStoryRegister(object):
output = pydub.AudioSegment(np.int16(audio * 2 ** 15).tobytes(), frame_rate=sample_rate, sample_width=2, channels=channels)
else:
output = output + pydub.AudioSegment(np.int16(audio * 2 ** 15).tobytes(), frame_rate=sample_rate, sample_width=2, channels=channels)
if output is not None:
output.export(filename, format="ogg", bitrate="16k")
def create_wave_slow(self, make_audio_queue_slow):
@@ -2123,7 +2124,7 @@ class KoboldStoryRegister(object):
sample_rate = 24000
speaker = 'train_daws'
if self.tortoise is None and importlib.util.find_spec("tortoise") is not None:
self.tortoise=api.TextToSpeech()
self.tortoise=api.TextToSpeech(use_deepspeed=os.environ.get('deepspeed', "true").lower()=="true", kv_cache=os.environ.get('kv_cache', "true").lower()=="true", half=True)
if importlib.util.find_spec("tortoise") is not None:
voice_samples, conditioning_latents = load_voices([speaker])
@@ -2135,8 +2136,17 @@ class KoboldStoryRegister(object):
if text.strip() == "":
shutil.copy("data/empty_audio.ogg", filename)
else:
if len(text) > 20000:
if len(self.tortoise.tokenizer.encode(text)) > 400:
text = self.sentence_re.findall(text)
i=0
while i <= len(text)-2:
if len(self.tortoise.tokenizer.encode(text[i] + text[i+1])) < 400:
text[i] = text[i] + text[i+1]
del text[i+1]
else:
i+=1
else:
text = [text]
output = None
@@ -2147,6 +2157,7 @@ class KoboldStoryRegister(object):
output = pydub.AudioSegment(np.int16(audio * 2 ** 15).tobytes(), frame_rate=sample_rate, sample_width=2, channels=channels)
else:
output = output + pydub.AudioSegment(np.int16(audio * 2 ** 15).tobytes(), frame_rate=sample_rate, sample_width=2, channels=channels)
if output is not None:
output.export(filename, format="ogg", bitrate="16k")
logger.info("Slow audio took {} for {} characters".format(time.time()-start_time, text_length))