Spaces:
Running
Running
mrfakename
commited on
Commit
•
68d3791
1
Parent(s):
ca0edb1
Update app_local.py
Browse files- app_local.py +79 -55
app_local.py
CHANGED
@@ -14,17 +14,17 @@ from pydub import AudioSegment
|
|
14 |
from model import CFM, UNetT, DiT, MMDiT
|
15 |
from cached_path import cached_path
|
16 |
from model.utils import (
|
17 |
-
get_tokenizer,
|
18 |
-
convert_char_to_pinyin,
|
19 |
save_spectrogram,
|
20 |
)
|
21 |
from transformers import pipeline
|
|
|
22 |
import librosa
|
|
|
23 |
|
24 |
device = "cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
|
25 |
|
26 |
-
print(f"Using {device} device")
|
27 |
-
|
28 |
pipe = pipeline(
|
29 |
"automatic-speech-recognition",
|
30 |
model="openai/whisper-large-v3-turbo",
|
@@ -79,13 +79,13 @@ E2TTS_model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)
|
|
79 |
F5TTS_ema_model, F5TTS_base_model = load_model("F5TTS_Base", DiT, F5TTS_model_cfg, 1200000)
|
80 |
E2TTS_ema_model, E2TTS_base_model = load_model("E2TTS_Base", UNetT, E2TTS_model_cfg, 1200000)
|
81 |
|
82 |
-
def infer(ref_audio_orig, ref_text, gen_text, exp_name, remove_silence):
|
83 |
print(gen_text)
|
84 |
-
if len(gen_text) > 200:
|
85 |
-
raise gr.Error("Please keep your text under 200 chars.")
|
86 |
gr.Info("Converting audio...")
|
87 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
|
88 |
aseg = AudioSegment.from_file(ref_audio_orig)
|
|
|
|
|
89 |
audio_duration = len(aseg)
|
90 |
if audio_duration > 15000:
|
91 |
gr.Warning("Audio is over 15s, clipping to only first 15s.")
|
@@ -98,7 +98,7 @@ def infer(ref_audio_orig, ref_text, gen_text, exp_name, remove_silence):
|
|
98 |
elif exp_name == "E2-TTS":
|
99 |
ema_model = E2TTS_ema_model
|
100 |
base_model = E2TTS_base_model
|
101 |
-
|
102 |
if not ref_text.strip():
|
103 |
gr.Info("No reference text provided, transcribing reference audio...")
|
104 |
ref_text = outputs = pipe(
|
@@ -112,7 +112,9 @@ def infer(ref_audio_orig, ref_text, gen_text, exp_name, remove_silence):
|
|
112 |
else:
|
113 |
gr.Info("Using custom reference text...")
|
114 |
audio, sr = torchaudio.load(ref_audio)
|
115 |
-
|
|
|
|
|
116 |
rms = torch.sqrt(torch.mean(torch.square(audio)))
|
117 |
if rms < target_rms:
|
118 |
audio = audio * target_rms / rms
|
@@ -120,44 +122,49 @@ def infer(ref_audio_orig, ref_text, gen_text, exp_name, remove_silence):
|
|
120 |
resampler = torchaudio.transforms.Resample(sr, target_sample_rate)
|
121 |
audio = resampler(audio)
|
122 |
audio = audio.to(device)
|
123 |
-
|
124 |
-
#
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
|
|
|
|
|
|
|
|
|
|
161 |
if remove_silence:
|
162 |
gr.Info("Removing audio silences... This may take a moment")
|
163 |
non_silent_intervals = librosa.effects.split(generated_wave, top_db=30)
|
@@ -169,11 +176,11 @@ def infer(ref_audio_orig, ref_text, gen_text, exp_name, remove_silence):
|
|
169 |
|
170 |
|
171 |
# spectogram
|
172 |
-
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp_spectrogram:
|
173 |
-
|
174 |
-
|
175 |
|
176 |
-
return (target_sample_rate, generated_wave)
|
177 |
|
178 |
with gr.Blocks() as app:
|
179 |
gr.Markdown("""
|
@@ -190,21 +197,38 @@ The checkpoints support English and Chinese.
|
|
190 |
|
191 |
If you're having issues, try converting your reference audio to WAV or MP3, clipping it to 15s, and shortening your prompt. If you're still running into issues, please open a [community Discussion](https://huggingface.co/spaces/mrfakename/E2-F5-TTS/discussions).
|
192 |
|
|
|
|
|
193 |
**NOTE: Reference text will be automatically transcribed with Whisper if not provided. For best results, keep your reference clips short (<15s). Ensure the audio is fully uploaded before generating.**
|
194 |
""")
|
195 |
|
196 |
ref_audio_input = gr.Audio(label="Reference Audio", type="filepath")
|
197 |
-
gen_text_input = gr.Textbox(label="Text to Generate (
|
198 |
model_choice = gr.Radio(choices=["F5-TTS", "E2-TTS"], label="Choose TTS Model", value="F5-TTS")
|
199 |
generate_btn = gr.Button("Synthesize", variant="primary")
|
200 |
with gr.Accordion("Advanced Settings", open=False):
|
201 |
ref_text_input = gr.Textbox(label="Reference Text", info="Leave blank to automatically transcribe the reference audio. If you enter text it will override automatic transcription.", lines=2)
|
202 |
remove_silence = gr.Checkbox(label="Remove Silences", info="The model tends to produce silences, especially on longer audio. We can manually remove silences if needed. Note that this is an experimental feature and may produce strange results. This will also increase generation time.", value=True)
|
203 |
-
|
204 |
audio_output = gr.Audio(label="Synthesized Audio")
|
205 |
-
spectrogram_output = gr.Image(label="Spectrogram")
|
|
|
|
|
|
|
|
|
|
|
|
|
206 |
|
207 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
208 |
gr.Markdown("Unofficial demo by [mrfakename](https://x.com/realmrfakename)")
|
209 |
|
210 |
|
|
|
14 |
from model import CFM, UNetT, DiT, MMDiT
|
15 |
from cached_path import cached_path
|
16 |
from model.utils import (
|
17 |
+
get_tokenizer,
|
18 |
+
convert_char_to_pinyin,
|
19 |
save_spectrogram,
|
20 |
)
|
21 |
from transformers import pipeline
|
22 |
+
import spaces
|
23 |
import librosa
|
24 |
+
from txtsplit import txtsplit
|
25 |
|
26 |
device = "cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"
|
27 |
|
|
|
|
|
28 |
pipe = pipeline(
|
29 |
"automatic-speech-recognition",
|
30 |
model="openai/whisper-large-v3-turbo",
|
|
|
79 |
F5TTS_ema_model, F5TTS_base_model = load_model("F5TTS_Base", DiT, F5TTS_model_cfg, 1200000)
|
80 |
E2TTS_ema_model, E2TTS_base_model = load_model("E2TTS_Base", UNetT, E2TTS_model_cfg, 1200000)
|
81 |
|
82 |
+
def infer(ref_audio_orig, ref_text, gen_text, exp_name, remove_silence, progress = gr.Progress()):
|
83 |
print(gen_text)
|
|
|
|
|
84 |
gr.Info("Converting audio...")
|
85 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
|
86 |
aseg = AudioSegment.from_file(ref_audio_orig)
|
87 |
+
# Convert to mono
|
88 |
+
aseg = aseg.set_channels(1)
|
89 |
audio_duration = len(aseg)
|
90 |
if audio_duration > 15000:
|
91 |
gr.Warning("Audio is over 15s, clipping to only first 15s.")
|
|
|
98 |
elif exp_name == "E2-TTS":
|
99 |
ema_model = E2TTS_ema_model
|
100 |
base_model = E2TTS_base_model
|
101 |
+
|
102 |
if not ref_text.strip():
|
103 |
gr.Info("No reference text provided, transcribing reference audio...")
|
104 |
ref_text = outputs = pipe(
|
|
|
112 |
else:
|
113 |
gr.Info("Using custom reference text...")
|
114 |
audio, sr = torchaudio.load(ref_audio)
|
115 |
+
# Audio
|
116 |
+
if audio.shape[0] > 1:
|
117 |
+
audio = torch.mean(audio, dim=0, keepdim=True)
|
118 |
rms = torch.sqrt(torch.mean(torch.square(audio)))
|
119 |
if rms < target_rms:
|
120 |
audio = audio * target_rms / rms
|
|
|
122 |
resampler = torchaudio.transforms.Resample(sr, target_sample_rate)
|
123 |
audio = resampler(audio)
|
124 |
audio = audio.to(device)
|
125 |
+
# Chunk
|
126 |
+
chunks = txtsplit(gen_text, 100, 150) # 100 chars preferred, 150 max
|
127 |
+
results = []
|
128 |
+
generated_mel_specs = []
|
129 |
+
for chunk in progress.tqdm(chunks):
|
130 |
+
# Prepare the text
|
131 |
+
text_list = [ref_text + chunk]
|
132 |
+
final_text_list = convert_char_to_pinyin(text_list)
|
133 |
+
|
134 |
+
# Calculate duration
|
135 |
+
ref_audio_len = audio.shape[-1] // hop_length
|
136 |
+
# if fix_duration is not None:
|
137 |
+
# duration = int(fix_duration * target_sample_rate / hop_length)
|
138 |
+
# else:
|
139 |
+
zh_pause_punc = r"。,、;:?!"
|
140 |
+
ref_text_len = len(ref_text) + len(re.findall(zh_pause_punc, ref_text))
|
141 |
+
gen_text_len = len(gen_text) + len(re.findall(zh_pause_punc, gen_text))
|
142 |
+
duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / speed)
|
143 |
+
|
144 |
+
# inference
|
145 |
+
gr.Info(f"Generating audio using {exp_name}")
|
146 |
+
with torch.inference_mode():
|
147 |
+
generated, _ = base_model.sample(
|
148 |
+
cond=audio,
|
149 |
+
text=final_text_list,
|
150 |
+
duration=duration,
|
151 |
+
steps=nfe_step,
|
152 |
+
cfg_strength=cfg_strength,
|
153 |
+
sway_sampling_coef=sway_sampling_coef,
|
154 |
+
)
|
155 |
+
|
156 |
+
generated = generated[:, ref_audio_len:, :]
|
157 |
+
generated_mel_spec = rearrange(generated, '1 n d -> 1 d n')
|
158 |
+
gr.Info("Running vocoder")
|
159 |
+
vocos = Vocos.from_pretrained("charactr/vocos-mel-24khz")
|
160 |
+
generated_wave = vocos.decode(generated_mel_spec.cpu())
|
161 |
+
if rms < target_rms:
|
162 |
+
generated_wave = generated_wave * rms / target_rms
|
163 |
+
|
164 |
+
# wav -> numpy
|
165 |
+
generated_wave = generated_wave.squeeze().cpu().numpy()
|
166 |
+
results.append(generated_wave)
|
167 |
+
generated_wave = np.concatenate(results)
|
168 |
if remove_silence:
|
169 |
gr.Info("Removing audio silences... This may take a moment")
|
170 |
non_silent_intervals = librosa.effects.split(generated_wave, top_db=30)
|
|
|
176 |
|
177 |
|
178 |
# spectogram
|
179 |
+
# with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp_spectrogram:
|
180 |
+
# spectrogram_path = tmp_spectrogram.name
|
181 |
+
# save_spectrogram(generated_mel_spec[0].cpu().numpy(), spectrogram_path)
|
182 |
|
183 |
+
return (target_sample_rate, generated_wave)
|
184 |
|
185 |
with gr.Blocks() as app:
|
186 |
gr.Markdown("""
|
|
|
197 |
|
198 |
If you're having issues, try converting your reference audio to WAV or MP3, clipping it to 15s, and shortening your prompt. If you're still running into issues, please open a [community Discussion](https://huggingface.co/spaces/mrfakename/E2-F5-TTS/discussions).
|
199 |
|
200 |
+
Long-form/batched inference + speech editing is coming soon!
|
201 |
+
|
202 |
**NOTE: Reference text will be automatically transcribed with Whisper if not provided. For best results, keep your reference clips short (<15s). Ensure the audio is fully uploaded before generating.**
|
203 |
""")
|
204 |
|
205 |
ref_audio_input = gr.Audio(label="Reference Audio", type="filepath")
|
206 |
+
gen_text_input = gr.Textbox(label="Text to Generate (longer text will use chunking)", lines=4)
|
207 |
model_choice = gr.Radio(choices=["F5-TTS", "E2-TTS"], label="Choose TTS Model", value="F5-TTS")
|
208 |
generate_btn = gr.Button("Synthesize", variant="primary")
|
209 |
with gr.Accordion("Advanced Settings", open=False):
|
210 |
ref_text_input = gr.Textbox(label="Reference Text", info="Leave blank to automatically transcribe the reference audio. If you enter text it will override automatic transcription.", lines=2)
|
211 |
remove_silence = gr.Checkbox(label="Remove Silences", info="The model tends to produce silences, especially on longer audio. We can manually remove silences if needed. Note that this is an experimental feature and may produce strange results. This will also increase generation time.", value=True)
|
212 |
+
|
213 |
audio_output = gr.Audio(label="Synthesized Audio")
|
214 |
+
# spectrogram_output = gr.Image(label="Spectrogram")
|
215 |
+
|
216 |
+
generate_btn.click(infer, inputs=[ref_audio_input, ref_text_input, gen_text_input, model_choice, remove_silence], outputs=[audio_output])
|
217 |
+
gr.Markdown("""
|
218 |
+
## Run Locally
|
219 |
+
|
220 |
+
Run this demo locally on CPU, CUDA, or MPS/Apple Silicon (requires macOS >= 14):
|
221 |
|
222 |
+
First, ensure `ffmpeg` is installed.
|
223 |
+
|
224 |
+
```bash
|
225 |
+
git clone https://huggingface.co/spaces/mrfakename/E2-F5-TTS
|
226 |
+
cd E2-F5-TTS
|
227 |
+
python -m pip install -r requirements.txt
|
228 |
+
python app_local.py
|
229 |
+
```
|
230 |
+
|
231 |
+
""")
|
232 |
gr.Markdown("Unofficial demo by [mrfakename](https://x.com/realmrfakename)")
|
233 |
|
234 |
|