Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -50,11 +50,35 @@ print("Loading WavLM for content...")
|
|
| 50 |
cmodel = WavLMModel.from_pretrained("microsoft/wavlm-large").to(device)
|
| 51 |
|
| 52 |
|
| 53 |
-
from openai import OpenAI
|
| 54 |
|
| 55 |
import ffmpeg
|
| 56 |
|
| 57 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
|
| 59 |
html_denoise = """
|
| 60 |
<html>
|
|
@@ -102,17 +126,10 @@ def convert(api_key, text, tgt, voice, save_path):
|
|
| 102 |
hps.data.mel_fmax
|
| 103 |
)
|
| 104 |
# src
|
| 105 |
-
client = OpenAI(api_key=api_key)
|
| 106 |
-
|
| 107 |
-
response = client.audio.speech.create(
|
| 108 |
-
model="tts-1",
|
| 109 |
-
voice=voice,
|
| 110 |
-
input=text,
|
| 111 |
-
)
|
| 112 |
-
|
| 113 |
-
response.stream_to_file("output_openai.mp3")
|
| 114 |
|
| 115 |
-
|
|
|
|
|
|
|
| 116 |
wav_src, _ = librosa.load(src, sr=hps.data.sampling_rate)
|
| 117 |
wav_src = torch.from_numpy(wav_src).unsqueeze(0).to(device)
|
| 118 |
c = cmodel(wav_src).last_hidden_state.transpose(1, 2).to(device)
|
|
|
|
| 50 |
cmodel = WavLMModel.from_pretrained("microsoft/wavlm-large").to(device)
|
| 51 |
|
| 52 |
|
|
|
|
| 53 |
|
| 54 |
import ffmpeg
|
| 55 |
|
| 56 |
+
import random
|
| 57 |
+
import numpy as np
|
| 58 |
+
from elevenlabs import voices, generate, set_api_key, UnauthenticatedRateLimitError
|
| 59 |
+
|
| 60 |
+
def pad_buffer(audio):
|
| 61 |
+
# Pad buffer to multiple of 2 bytes
|
| 62 |
+
buffer_size = len(audio)
|
| 63 |
+
element_size = np.dtype(np.int16).itemsize
|
| 64 |
+
if buffer_size % element_size != 0:
|
| 65 |
+
audio = audio + b'\0' * (element_size - (buffer_size % element_size))
|
| 66 |
+
return audio
|
| 67 |
+
|
| 68 |
+
def generate_voice(text, voice_name):
|
| 69 |
+
try:
|
| 70 |
+
audio = generate(
|
| 71 |
+
text[:250], # Limit to 250 characters
|
| 72 |
+
voice=voice_name,
|
| 73 |
+
model="eleven_multilingual_v2"
|
| 74 |
+
)
|
| 75 |
+
write("output.wav", 44100, np.frombuffer(pad_buffer(audio), dtype=np.int16))
|
| 76 |
+
|
| 77 |
+
return "output.wav"
|
| 78 |
+
except UnauthenticatedRateLimitError as e:
|
| 79 |
+
raise gr.Error("Thanks for trying out ElevenLabs TTS! You've reached the free tier limit. Please provide an API key to continue.")
|
| 80 |
+
except Exception as e:
|
| 81 |
+
raise gr.Error(e)
|
| 82 |
|
| 83 |
html_denoise = """
|
| 84 |
<html>
|
|
|
|
| 126 |
hps.data.mel_fmax
|
| 127 |
)
|
| 128 |
# src
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 129 |
|
| 130 |
+
#response.stream_to_file("output_openai.mp3")
|
| 131 |
+
os.environ["ELEVEN_API_KEY"] = api_key
|
| 132 |
+
src = generate_voice(text, voice)
|
| 133 |
wav_src, _ = librosa.load(src, sr=hps.data.sampling_rate)
|
| 134 |
wav_src = torch.from_numpy(wav_src).unsqueeze(0).to(device)
|
| 135 |
c = cmodel(wav_src).last_hidden_state.transpose(1, 2).to(device)
|