mirrors/openbmb--voxcpm
1
0
Fork 0
Code Issues Pull Requests Actions Deployments Models Agent Notes Packages Projects Releases Wiki Activity

fix: formatting/trailing whitespace

Browse Source
This commit is contained in:
Eliseu Silva
2026-06-24 22:31:23 -03:00
parent 5ef0b3db4c
commit 5e293dee2f
16 changed files with 174 additions and 212 deletions
Download Patch File Download Diff File Expand all files Collapse all files
app.py
+12 -9
View File
@@ -45,8 +45,8 @@ _EXAMPLES_FOOTER_EN = (
"**Example 1 — Gentle & Melancholic Girl** \n"
'`Control Instruction`: *"A young girl with a soft, sweet voice. '
'Speaks slowly with a melancholic, slightly tsundere tone."* \n'
'`Target Text`: *"I never asked you to stay… It\'s not like I care or anything. '
'But… why does it still hurt so much now that you\'re gone?"* \n\n'
"`Target Text`: *\"I never asked you to stay… It's not like I care or anything. "
"But… why does it still hurt so much now that you're gone?\"* \n\n"
"**Example 2 — Laid-Back Surfer Dude** \n"
'`Control Instruction`: *"Relaxed young male voice, slightly nasal, '
'lazy drawl, very casual and chill."* \n'
@@ -147,7 +147,7 @@ _I18N_TRANSLATIONS = {
"examples_footer": _EXAMPLES_FOOTER_ZH,
},
"zh-Hans": None, # alias, filled below
"zh": None, # alias, filled below
"zh": None, # alias, filled below
}
_I18N_TRANSLATIONS["zh-Hans"] = _I18N_TRANSLATIONS["zh-CN"]
_I18N_TRANSLATIONS["zh"] = _I18N_TRANSLATIONS["zh-CN"]
@@ -160,8 +160,7 @@ for _d in _I18N_TRANSLATIONS.values():
I18N = gr.I18n(**_I18N_TRANSLATIONS)
DEFAULT_TARGET_TEXT = (
"VoxCPM2 is a creative multilingual TTS model from ModelBest, "
"designed to generate highly realistic speech."
"VoxCPM2 is a creative multilingual TTS model from ModelBest, " "designed to generate highly realistic speech."
)
_CUSTOM_CSS = """
@@ -224,6 +223,7 @@ _APP_THEME = gr.themes.Soft(
# ---------- Model ----------
class VoxCPMDemo:
def __init__(self, model_id: str = "openbmb/VoxCPM2", device: str = "auto") -> None:
self.device = resolve_runtime_device(device, "cuda")
@@ -252,9 +252,7 @@ class VoxCPMDemo:
def get_or_load_asr_model(self) -> AutoModel:
if self.asr_model is not None:
return self.asr_model
logger.info(
f"Loading ASR model: {self.asr_model_id} on device: {self.asr_device}"
)
logger.info(f"Loading ASR model: {self.asr_model_id} on device: {self.asr_device}")
self.asr_model = AutoModel(
model=self.asr_model_id,
disable_update=True,
@@ -352,6 +350,7 @@ class VoxCPMDemo:
# ---------- UI ----------
def create_demo_interface(demo: VoxCPMDemo):
gr.set_static_paths(paths=[Path.cwd().absolute() / "assets"])
@@ -554,6 +553,7 @@ def create_demo_interface(demo: VoxCPMDemo):
return interface
def run_demo(
server_name: str = "0.0.0.0",
server_port: int = 8808,
@@ -575,9 +575,12 @@ def run_demo(
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser()
parser.add_argument(
"--model-id", type=str, default="openbmb/VoxCPM2",
"--model-id",
type=str,
default="openbmb/VoxCPM2",
help="Local path or HuggingFace repo ID (default: openbmb/VoxCPM2)",
)
parser.add_argument("--port", type=int, default=8808, help="Server port")
app_old.py
+24 -12
View File
@@ -6,6 +6,7 @@ import gradio as gr
from typing import Optional, Tuple
from funasr import AutoModel
from pathlib import Path
os.environ["TOKENIZERS_PARALLELISM"] = "false"
if os.environ.get("HF_REPO_ID", "").strip() == "":
os.environ["HF_REPO_ID"] = "openbmb/VoxCPM1.5"
@@ -23,7 +24,7 @@ class VoxCPMDemo:
self.asr_model: Optional[AutoModel] = AutoModel(
model=self.asr_model_id,
disable_update=True,
log_level='DEBUG',
log_level="DEBUG",
device="cuda:0" if self.device == "cuda" else "cpu",
)
@@ -48,6 +49,7 @@ class VoxCPMDemo:
if not os.path.isdir(target_dir):
try:
from huggingface_hub import snapshot_download # type: ignore
os.makedirs(target_dir, exist_ok=True)
print(f"Downloading model from HF repo '{repo_id}' to '{target_dir}' ...", file=sys.stderr)
snapshot_download(repo_id=repo_id, local_dir=target_dir, local_dir_use_symlinks=False)
@@ -72,7 +74,7 @@ class VoxCPMDemo:
if prompt_wav is None:
return ""
res = self.asr_model.generate(input=prompt_wav, language="auto", use_itn=True)
text = res[0]["text"].split('|>')[-1]
text = res[0]["text"].split("|>")[-1]
return text
def generate_tts_audio(
@@ -149,11 +151,13 @@ _CUSTOM_CSS = """
def create_demo_interface(demo: VoxCPMDemo):
"""Build the Gradio UI for VoxCPM demo."""
gr.set_static_paths(paths=[Path.cwd().absolute()/"assets"])
gr.set_static_paths(paths=[Path.cwd().absolute() / "assets"])
with gr.Blocks() as interface:
# Header logo
gr.HTML('<div class="logo-container"><img src="/gradio_api/file=assets/voxcpm_logo.png" alt="VoxCPM Logo"></div>')
gr.HTML(
'<div class="logo-container"><img src="/gradio_api/file=assets/voxcpm_logo.png" alt="VoxCPM Logo"></div>'
)
# Quick Start
with gr.Accordion("📋 Quick Start Guide ｜快速入门", open=False, elem_id="acc_quick"):
@@ -201,7 +205,7 @@ def create_demo_interface(demo: VoxCPMDemo):
with gr.Row():
with gr.Column():
prompt_wav = gr.Audio(
sources=["upload", 'microphone'],
sources=["upload", "microphone"],
type="filepath",
label="Prompt Speech (Optional, or let VoxCPM improvise)",
value="./examples/example.wav",
@@ -210,13 +214,13 @@ def create_demo_interface(demo: VoxCPMDemo):
value=False,
label="Prompt Speech Enhancement",
elem_id="chk_denoise",
info="We use ZipEnhancer model to denoise the prompt audio."
info="We use ZipEnhancer model to denoise the prompt audio.",
)
with gr.Row():
prompt_text = gr.Textbox(
value="Just by listening a few minutes a day, you'll be able to eliminate negative thoughts by conditioning your mind to be more positive.",
label="Prompt Text",
placeholder="Please enter the prompt text. Automatic recognition is supported, and you can correct the results yourself..."
placeholder="Please enter the prompt text. Automatic recognition is supported, and you can correct the results yourself...",
)
run_btn = gr.Button("Generate Speech", variant="primary")
@@ -227,7 +231,7 @@ def create_demo_interface(demo: VoxCPMDemo):
value=2.0,
step=0.1,
label="CFG Value (Guidance Scale)",
info="Higher values increase adherence to prompt, lower values allow more creativity"
info="Higher values increase adherence to prompt, lower values allow more creativity",
)
inference_timesteps = gr.Slider(
minimum=4,
@@ -235,7 +239,7 @@ def create_demo_interface(demo: VoxCPMDemo):
value=10,
step=1,
label="Inference Timesteps",
info="Number of inference timesteps for generation (higher values may improve quality but slower)"
info="Number of inference timesteps for generation (higher values may improve quality but slower)",
)
with gr.Row():
text = gr.Textbox(
@@ -247,14 +251,22 @@ def create_demo_interface(demo: VoxCPMDemo):
value=False,
label="Text Normalization",
elem_id="chk_normalize",
info="We use wetext library to normalize the input text."
info="We use wetext library to normalize the input text.",
)
audio_output = gr.Audio(label="Output Audio")
# Wiring
run_btn.click(
fn=demo.generate_tts_audio,
inputs=[text, prompt_wav, prompt_text, cfg_value, inference_timesteps, DoNormalizeText, DoDenoisePromptAudio],
inputs=[
text,
prompt_wav,
prompt_text,
cfg_value,
inference_timesteps,
DoNormalizeText,
DoDenoisePromptAudio,
],
outputs=[audio_output],
show_progress=True,
api_name="generate",
@@ -277,4 +289,4 @@ def run_demo(server_name: str = "localhost", server_port: int = 7860, show_error
if __name__ == "__main__":
run_demo()
run_demo()
lora_ft_webui.py
+3 -4
View File
@@ -308,7 +308,8 @@ def run_inference(text, prompt_wav, prompt_text, lora_selection, cfg_scale, step
if new_r is not None and current_r is not None and new_r != current_r:
print(f"LoRA rank mismatch (model r={current_r}, checkpoint r={new_r}), reloading...", file=sys.stderr)
reload_base = (
new_base_model if new_base_model and os.path.exists(new_base_model)
new_base_model
if new_base_model and os.path.exists(new_base_model)
else (pretrained_path if pretrained_path and pretrained_path.strip() else default_pretrained_path)
)
try:
@@ -982,9 +983,7 @@ with gr.Blocks(title="VoxCPM LoRA WebUI", theme=gr.themes.Soft(), css=custom_css
gr.Markdown("#### 分发选项 (Distribution)")
with gr.Row():
hf_model_id = gr.Textbox(
label="HuggingFace Model ID (e.g., openbmb/VoxCPM2)", value=""
)
hf_model_id = gr.Textbox(label="HuggingFace Model ID (e.g., openbmb/VoxCPM2)", value="")
distribute = gr.Checkbox(label="分发模式 (distribute)", value=False)
with gr.Column(scale=2, elem_classes="form-section"):
scripts/test_pick_runtime_dtype.py
+1
View File
@@ -3,6 +3,7 @@
Loads src/voxcpm/model/utils.py directly to avoid the heavy voxcpm package
init. Run with: `python scripts/test_pick_runtime_dtype.py`.
"""
import importlib.util
import os
import pathlib
scripts/test_voxcpm_ft_infer.py
+1 -1
View File
@@ -143,4 +143,4 @@ def main():
if __name__ == "__main__":
main()
main()
scripts/test_voxcpm_lora_infer.py
+1 -1
View File
@@ -273,4 +273,4 @@ def main():
if __name__ == "__main__":
main()
main()
scripts/train_voxcpm_finetune.py
+3 -1
View File
@@ -599,7 +599,9 @@ def generate_sample_audio(
)
with torch.no_grad():
with autocast_ctx:
generated = unwrapped_model.generate(target_text=text, inference_timesteps=10, cfg_value=2.0, seed=42 )
generated = unwrapped_model.generate(
target_text=text, inference_timesteps=10, cfg_value=2.0, seed=42
)
# Restore training setup
# unwrapped_model.to(torch.float32)
src/voxcpm/cli.py
+36 -109
View File
@@ -86,9 +86,7 @@ def resolve_prompt_text(args, parser) -> str | None:
def detect_model_architecture(args) -> str | None:
model_location = getattr(args, "model_path", None) or getattr(
args, "hf_model_id", None
)
model_location = getattr(args, "model_path", None) or getattr(args, "hf_model_id", None)
if not model_location:
return None
@@ -103,11 +101,7 @@ def detect_model_architecture(args) -> str | None:
model_hint = str(model_location).lower()
if "voxcpm2" in model_hint:
return "voxcpm2"
if (
"voxcpm1.5" in model_hint
or "voxcpm-1.5" in model_hint
or "voxcpm_1.5" in model_hint
):
if "voxcpm1.5" in model_hint or "voxcpm-1.5" in model_hint or "voxcpm_1.5" in model_hint:
return "voxcpm"
return None
@@ -121,9 +115,7 @@ def validate_prompt_related_args(args, parser, prompt_text: str | None):
parser.error("--prompt-audio requires --prompt-text or --prompt-file.")
if args.control and prompt_text:
parser.error(
"--control cannot be used together with --prompt-text or --prompt-file."
)
parser.error("--control cannot be used together with --prompt-text or --prompt-file.")
def validate_reference_support(args, parser):
@@ -138,9 +130,7 @@ def validate_reference_support(args, parser):
def validate_design_args(args, parser):
prompt_text = resolve_prompt_text(args, parser)
if args.prompt_audio or args.reference_audio or prompt_text:
parser.error(
"`design` does not accept prompt/reference audio. Use `clone` instead."
)
parser.error("`design` does not accept prompt/reference audio. Use `clone` instead.")
def validate_clone_args(args, parser):
@@ -149,9 +139,7 @@ def validate_clone_args(args, parser):
validate_reference_support(args, parser)
if not args.prompt_audio and not args.reference_audio:
parser.error(
"`clone` requires --reference-audio, or --prompt-audio with --prompt-text/--prompt-file."
)
parser.error("`clone` requires --reference-audio, or --prompt-audio with --prompt-text/--prompt-file.")
return prompt_text
@@ -173,9 +161,7 @@ def load_model(args):
print("Loading VoxCPM model...", file=sys.stderr)
zipenhancer_path = getattr(args, "zipenhancer_path", None) or os.environ.get(
"ZIPENHANCER_MODEL_PATH", None
)
zipenhancer_path = getattr(args, "zipenhancer_path", None) or os.environ.get("ZIPENHANCER_MODEL_PATH", None)
# Build LoRA config if provided
lora_config = None
@@ -259,8 +245,7 @@ def _run_single(args, parser, *, text: str, output: str, prompt_text: str | None
cfg_value=args.cfg_value,
inference_timesteps=args.inference_timesteps,
normalize=args.normalize,
denoise=args.denoise
and (args.prompt_audio is not None or args.reference_audio is not None),
denoise=args.denoise and (args.prompt_audio is not None or args.reference_audio is not None),
seed=args.seed,
)
@@ -275,17 +260,13 @@ def _run_single(args, parser, *, text: str, output: str, prompt_text: str | None
def cmd_design(args, parser):
validate_design_args(args, parser)
final_text = build_final_text(args.text, args.control)
return _run_single(
args, parser, text=final_text, output=args.output, prompt_text=None
)
return _run_single(args, parser, text=final_text, output=args.output, prompt_text=None)
def cmd_clone(args, parser):
prompt_text = validate_clone_args(args, parser)
final_text = build_final_text(args.text, args.control)
return _run_single(
args, parser, text=final_text, output=args.output, prompt_text=prompt_text
)
return _run_single(args, parser, text=final_text, output=args.output, prompt_text=prompt_text)
def cmd_validate(args, parser):
@@ -324,15 +305,11 @@ def cmd_batch(args, parser):
prompt_audio_path = None
if args.prompt_audio:
prompt_audio_path = str(
require_file_exists(args.prompt_audio, parser, "prompt audio file")
)
prompt_audio_path = str(require_file_exists(args.prompt_audio, parser, "prompt audio file"))
reference_audio_path = None
if args.reference_audio:
reference_audio_path = str(
require_file_exists(args.reference_audio, parser, "reference audio file")
)
reference_audio_path = str(require_file_exists(args.reference_audio, parser, "reference audio file"))
success_count = 0
@@ -347,8 +324,7 @@ def cmd_batch(args, parser):
cfg_value=args.cfg_value,
inference_timesteps=args.inference_timesteps,
normalize=args.normalize,
denoise=args.denoise
and (prompt_audio_path is not None or reference_audio_path is not None),
denoise=args.denoise and (prompt_audio_path is not None or reference_audio_path is not None),
seed=args.seed,
)
@@ -389,9 +365,7 @@ def _add_common_generation_args(parser):
default=10,
help="Inference steps (int, recommended 4–30, default: 10)",
)
parser.add_argument(
"--normalize", action="store_true", help="Enable text normalization"
)
parser.add_argument("--normalize", action="store_true", help="Enable text normalization")
parser.add_argument(
"--seed",
type=int,
@@ -406,12 +380,8 @@ def _add_prompt_reference_args(parser):
"-pa",
help="Prompt audio file path (continuation mode, requires --prompt-text or --prompt-file)",
)
parser.add_argument(
"--prompt-text", "-pt", help="Text corresponding to the prompt audio"
)
parser.add_argument(
"--prompt-file", type=str, help="Text file corresponding to the prompt audio"
)
parser.add_argument("--prompt-text", "-pt", help="Text corresponding to the prompt audio")
parser.add_argument("--prompt-file", type=str, help="Text file corresponding to the prompt audio")
parser.add_argument(
"--reference-audio",
"-ra",
@@ -438,15 +408,9 @@ def _add_model_args(parser):
default="auto",
help="Runtime device: auto, cpu, mps, cuda, or cuda:N (default: auto)",
)
parser.add_argument(
"--cache-dir", type=str, help="Cache directory for Hub downloads"
)
parser.add_argument(
"--local-files-only", action="store_true", help="Disable network access"
)
parser.add_argument(
"--no-denoiser", action="store_true", help="Disable denoiser model loading"
)
parser.add_argument("--cache-dir", type=str, help="Cache directory for Hub downloads")
parser.add_argument("--local-files-only", action="store_true", help="Disable network access")
parser.add_argument("--no-denoiser", action="store_true", help="Disable denoiser model loading")
parser.add_argument(
"--no-optimize",
action="store_true",
@@ -461,9 +425,7 @@ def _add_model_args(parser):
def _add_lora_args(parser):
parser.add_argument("--lora-path", type=str, help="Path to LoRA weights")
parser.add_argument(
"--lora-r", type=int, default=32, help="LoRA rank (positive int, default: 32)"
)
parser.add_argument("--lora-r", type=int, default=32, help="LoRA rank (positive int, default: 32)")
parser.add_argument(
"--lora-alpha",
type=int,
@@ -476,12 +438,8 @@ def _add_lora_args(parser):
default=0.0,
help="LoRA dropout rate (0.0–1.0, default: 0.0)",
)
parser.add_argument(
"--lora-disable-lm", action="store_true", help="Disable LoRA on LM layers"
)
parser.add_argument(
"--lora-disable-dit", action="store_true", help="Disable LoRA on DiT layers"
)
parser.add_argument("--lora-disable-lm", action="store_true", help="Disable LoRA on LM layers")
parser.add_argument("--lora-disable-dit", action="store_true", help="Disable LoRA on DiT layers")
parser.add_argument(
"--lora-enable-proj",
action="store_true",
@@ -504,37 +462,23 @@ Examples:
subparsers = parser.add_subparsers(dest="command")
design_parser = subparsers.add_parser(
"design", help="Generate speech with VoxCPM2-first voice design"
)
design_parser = subparsers.add_parser("design", help="Generate speech with VoxCPM2-first voice design")
_add_common_generation_args(design_parser)
_add_prompt_reference_args(design_parser)
_add_model_args(design_parser)
_add_lora_args(design_parser)
design_parser.add_argument(
"--output", "-o", required=True, help="Output audio file path"
)
design_parser.add_argument("--output", "-o", required=True, help="Output audio file path")
clone_parser = subparsers.add_parser(
"clone", help="Clone a voice with reference/prompt audio"
)
clone_parser = subparsers.add_parser("clone", help="Clone a voice with reference/prompt audio")
_add_common_generation_args(clone_parser)
_add_prompt_reference_args(clone_parser)
_add_model_args(clone_parser)
_add_lora_args(clone_parser)
clone_parser.add_argument(
"--output", "-o", required=True, help="Output audio file path"
)
clone_parser.add_argument("--output", "-o", required=True, help="Output audio file path")
batch_parser = subparsers.add_parser(
"batch", help="Batch-generate one line per output file"
)
batch_parser.add_argument(
"--input", "-i", required=True, help="Input text file (one text per line)"
)
batch_parser.add_argument(
"--output-dir", "-od", required=True, help="Output directory"
)
batch_parser = subparsers.add_parser("batch", help="Batch-generate one line per output file")
batch_parser.add_argument("--input", "-i", required=True, help="Input text file (one text per line)")
batch_parser.add_argument("--output-dir", "-od", required=True, help="Output directory")
batch_parser.add_argument(
"--control",
type=str,
@@ -553,9 +497,7 @@ Examples:
default=10,
help="Inference steps (int, recommended 4–30, default: 10)",
)
batch_parser.add_argument(
"--normalize", action="store_true", help="Enable text normalization"
)
batch_parser.add_argument("--normalize", action="store_true", help="Enable text normalization")
batch_parser.add_argument(
"--seed",
type=int,
@@ -570,9 +512,7 @@ Examples:
"validate",
help="Validate a training data manifest (JSONL) before fine-tuning",
)
validate_parser.add_argument(
"--manifest", "-m", required=True, help="Path to JSONL training manifest"
)
validate_parser.add_argument("--manifest", "-m", required=True, help="Path to JSONL training manifest")
validate_parser.add_argument(
"--sample-rate",
type=int,
@@ -585,19 +525,13 @@ Examples:
default=0,
help="Maximum number of samples to validate (0 = all, default: 0)",
)
validate_parser.add_argument(
"--verbose", "-v", action="store_true", help="Print per-sample progress"
)
validate_parser.add_argument("--verbose", "-v", action="store_true", help="Print per-sample progress")
# Legacy root arguments
parser.add_argument("--input", "-i", help="Input text file (batch mode only)")
parser.add_argument(
"--output-dir", "-od", help="Output directory (batch mode only)"
)
parser.add_argument("--output-dir", "-od", help="Output directory (batch mode only)")
_add_common_generation_args(parser)
parser.add_argument(
"--output", "-o", help="Output audio file path (single or clone mode)"
)
parser.add_argument("--output", "-o", help="Output audio file path (single or clone mode)")
_add_prompt_reference_args(parser)
_add_model_args(parser)
_add_lora_args(parser)
@@ -609,9 +543,7 @@ def _dispatch_legacy(args, parser):
warn_legacy_mode()
if args.input and args.text:
parser.error(
"Use either batch mode (--input) or single mode (--text), not both."
)
parser.error("Use either batch mode (--input) or single mode (--text), not both.")
if args.input:
if not args.output_dir:
@@ -621,12 +553,7 @@ def _dispatch_legacy(args, parser):
if not args.text or not args.output:
parser.error("Single-sample legacy mode requires --text and --output")
if (
args.prompt_audio
or args.prompt_text
or args.prompt_file
or args.reference_audio
):
if args.prompt_audio or args.prompt_text or args.prompt_file or args.reference_audio:
return cmd_clone(args, parser)
return cmd_design(args, parser)
@@ -663,4 +590,4 @@ def main():
if __name__ == "__main__":
main()
main()
src/voxcpm/core.py
+1 -1
View File
@@ -293,7 +293,7 @@ class VoxCPM:
retry_badcase_max_times=retry_badcase_max_times,
retry_badcase_ratio_threshold=retry_badcase_ratio_threshold,
streaming=streaming,
seed=seed
seed=seed,
)
if streaming:
src/voxcpm/model/utils.py
+9 -11
View File
@@ -5,9 +5,12 @@ from transformers import PreTrainedTokenizer
_LOW_PRECISION_DTYPES = {"bfloat16", "bf16", "float16", "fp16"}
_VALID_DTYPE_OVERRIDES = {
"bfloat16", "bf16",
"float16", "fp16",
"float32", "fp32",
"bfloat16",
"bf16",
"float16",
"fp16",
"float32",
"fp32",
}
@@ -173,10 +176,7 @@ def pick_runtime_dtype(device: str, configured_dtype: str) -> str:
override = os.environ.get("VOXCPM_MPS_DTYPE", "").strip().lower()
if override:
if override not in _VALID_DTYPE_OVERRIDES:
raise ValueError(
f"VOXCPM_MPS_DTYPE='{override}' is not one of "
f"{sorted(_VALID_DTYPE_OVERRIDES)}"
)
raise ValueError(f"VOXCPM_MPS_DTYPE='{override}' is not one of " f"{sorted(_VALID_DTYPE_OVERRIDES)}")
return override
if (configured_dtype or "").lower() in _LOW_PRECISION_DTYPES:
@@ -224,15 +224,13 @@ def resolve_runtime_device(device: Optional[str], configured_device: str = "cuda
if explicit.startswith("cuda"):
if not torch.cuda.is_available():
raise ValueError(
f"Requested device '{device}', but CUDA is not available. "
"Use device='auto' for automatic fallback."
f"Requested device '{device}', but CUDA is not available. " "Use device='auto' for automatic fallback."
)
return explicit
if explicit == "mps":
if not _has_mps():
raise ValueError(
"Requested device 'mps', but MPS is not available. "
"Use device='auto' for automatic fallback."
"Requested device 'mps', but MPS is not available. " "Use device='auto' for automatic fallback."
)
return "mps"
if explicit == "cpu":
src/voxcpm/model/voxcpm2.py
+5 -3
View File
@@ -57,7 +57,9 @@ from .utils import (
# A simple function to trim audio silence using VAD, not used default
def _trim_audio_silence_vad(audio: torch.Tensor, sample_rate: int, max_silence_ms: float = 200.0, top_db: float = 35.0) -> torch.Tensor:
def _trim_audio_silence_vad(
audio: torch.Tensor, sample_rate: int, max_silence_ms: float = 200.0, top_db: float = 35.0
) -> torch.Tensor:
if audio.numel() == 0:
return audio
y = audio.squeeze(0).numpy()
@@ -684,7 +686,7 @@ class VoxCPM2Model(nn.Module):
decode_audio = self.audio_vae.decode(latent_pred.to(torch.float32))
decode_patch_len = self.patch_size * self._decode_chunk_size
if context_len > 0:
decode_audio = decode_audio[..., decode_patch_len * context_len:].squeeze(1).cpu()
decode_audio = decode_audio[..., decode_patch_len * context_len :].squeeze(1).cpu()
else:
decode_audio = decode_audio.squeeze(1).cpu()
yield decode_audio
@@ -979,7 +981,7 @@ class VoxCPM2Model(nn.Module):
decode_audio = self.audio_vae.decode(latent_pred.to(torch.float32))
decode_patch_len = self.patch_size * self._decode_chunk_size
if context_len > 0:
decode_audio = decode_audio[..., decode_patch_len * context_len:].squeeze(1).cpu()
decode_audio = decode_audio[..., decode_patch_len * context_len :].squeeze(1).cpu()
else:
decode_audio = decode_audio.squeeze(1).cpu()
yield (decode_audio, target_text_token, pred_audio_feat)
src/voxcpm/modules/audiovae/audio_vae_v2.py
+1 -2
View File
@@ -551,8 +551,7 @@ class StreamingVAEDecoder:
if x.shape[-1] >= _p:
states[_k] = x[:, :, -_p:].detach()
else:
prev = states.get(_k, torch.zeros(x.shape[0], x.shape[1], _p,
device=x.device, dtype=x.dtype))
prev = states.get(_k, torch.zeros(x.shape[0], x.shape[1], _p, device=x.device, dtype=x.dtype))
states[_k] = torch.cat([prev, x], dim=-1)[:, :, -_p:].detach()
return nn.Conv1d.forward(_m, x_pad)
src/voxcpm/training/packers.py
+66 -31
View File
@@ -350,45 +350,80 @@ class AudioFeatureProcessingPacker:
# -- text token track --
# [103, 0×R, 104, text_ids, 101, 0×A, 102]
text_token_info = torch.cat([
_tok([self.audio_prompt_start_id]),
torch.zeros(ref_len, dtype=torch.int32, device=device),
_tok([self.audio_prompt_end_id]),
text_token,
_tok([self.audio_start_id]),
torch.zeros(tgt_len, dtype=torch.int32, device=device),
_tok([self.audio_end_id]),
])
text_token_info = torch.cat(
[
_tok([self.audio_prompt_start_id]),
torch.zeros(ref_len, dtype=torch.int32, device=device),
_tok([self.audio_prompt_end_id]),
text_token,
_tok([self.audio_start_id]),
torch.zeros(tgt_len, dtype=torch.int32, device=device),
_tok([self.audio_end_id]),
]
)
# -- audio feature track --
zero_1 = torch.zeros((1,) + feat_shape, dtype=torch.float32, device=device)
zero_txt = torch.zeros((txt_len,) + feat_shape, dtype=torch.float32, device=device)
audio_feat_info = torch.cat([
zero_1, ref_feats, zero_1, # 103, ref, 104
zero_txt, # text
zero_1, tgt_feats, zero_1, # 101, target, 102
], dim=0)
audio_feat_info = torch.cat(
[
zero_1,
ref_feats,
zero_1, # 103, ref, 104
zero_txt, # text
zero_1,
tgt_feats,
zero_1, # 101, target, 102
],
dim=0,
)
# -- masks --
text_mask = torch.cat([
torch.ones(1), torch.zeros(ref_len), torch.ones(1),
torch.ones(txt_len),
torch.ones(1), torch.zeros(tgt_len), torch.ones(1),
]).to(torch.int32).to(device)
text_mask = (
torch.cat(
[
torch.ones(1),
torch.zeros(ref_len),
torch.ones(1),
torch.ones(txt_len),
torch.ones(1),
torch.zeros(tgt_len),
torch.ones(1),
]
)
.to(torch.int32)
.to(device)
)
audio_mask = torch.cat([
torch.zeros(1), torch.ones(ref_len), torch.zeros(1),
torch.zeros(txt_len),
torch.zeros(1), torch.ones(tgt_len), torch.zeros(1),
]).to(torch.int32).to(device)
audio_mask = (
torch.cat(
[
torch.zeros(1),
torch.ones(ref_len),
torch.zeros(1),
torch.zeros(txt_len),
torch.zeros(1),
torch.ones(tgt_len),
torch.zeros(1),
]
)
.to(torch.int32)
.to(device)
)
loss_mask = torch.cat([
torch.zeros(1 + ref_len + 1), # ref part: no loss
torch.zeros(txt_len), # text: no loss
torch.zeros(1), # 101: no loss
torch.ones(tgt_len), # target audio: LOSS
torch.zeros(1), # 102: no loss
]).to(torch.int32).to(device)
loss_mask = (
torch.cat(
[
torch.zeros(1 + ref_len + 1), # ref part: no loss
torch.zeros(txt_len), # text: no loss
torch.zeros(1), # 101: no loss
torch.ones(tgt_len), # target audio: LOSS
torch.zeros(1), # 102: no loss
]
)
.to(torch.int32)
.to(device)
)
total_len = 1 + ref_len + 1 + txt_len + 1 + tgt_len + 1
labels = torch.zeros(total_len, dtype=torch.int32, device=device)
src/voxcpm/training/validate.py
+6 -19
View File
@@ -44,10 +44,7 @@ def _check_audio_file(audio_path: str, sample_rate: int) -> Optional[str]:
if info.frames == 0:
return f"Audio file is empty: {audio_path}"
if info.samplerate != sample_rate:
return (
f"Sample rate mismatch in {audio_path}: "
f"expected {sample_rate} Hz, got {info.samplerate} Hz"
)
return f"Sample rate mismatch in {audio_path}: " f"expected {sample_rate} Hz, got {info.samplerate} Hz"
return None
except ImportError:
# soundfile not available; just check existence
@@ -187,13 +184,9 @@ def validate_manifest(
if duration is not None:
result.audio_durations.append(duration)
if duration < 0.3:
result.warnings.append(
f"Line {i + 1}: Very short audio ({duration:.2f}s)"
)
result.warnings.append(f"Line {i + 1}: Very short audio ({duration:.2f}s)")
elif duration > 30.0:
result.warnings.append(
f"Line {i + 1}: Very long audio ({duration:.1f}s), may cause OOM"
)
result.warnings.append(f"Line {i + 1}: Very long audio ({duration:.1f}s), may cause OOM")
else:
result.errors.append(f"Line {i + 1}: Invalid audio path")
has_error = True
@@ -209,9 +202,7 @@ def validate_manifest(
if os.path.isfile(ref_path):
result.has_ref_audio += 1
else:
result.warnings.append(
f"Line {i + 1}: ref_audio file not found: {ref_path}"
)
result.warnings.append(f"Line {i + 1}: ref_audio file not found: {ref_path}")
if not has_error:
result.valid_samples += 1
@@ -222,14 +213,10 @@ def validate_manifest(
# Summarize truncated errors
if missing_audio_count > 5:
result.errors.append(
f"... and {missing_audio_count - 5} more missing audio files "
f"({missing_audio_count} total)"
f"... and {missing_audio_count - 5} more missing audio files " f"({missing_audio_count} total)"
)
if empty_text_count > 5:
result.warnings.append(
f"... and {empty_text_count - 5} more empty text entries "
f"({empty_text_count} total)"
)
result.warnings.append(f"... and {empty_text_count - 5} more empty text entries " f"({empty_text_count} total)")
return result
tests/test_cli.py
+3 -8
View File
@@ -455,10 +455,7 @@ def test_clone_rejects_prompt_audio_without_transcript(monkeypatch, tmp_path, ca
with pytest.raises(SystemExit):
cli.main()
assert (
"--prompt-audio requires --prompt-text or --prompt-file"
in capsys.readouterr().err
)
assert "--prompt-audio requires --prompt-text or --prompt-file" in capsys.readouterr().err
def test_clone_rejects_transcript_without_prompt_audio(monkeypatch, tmp_path, capsys):
@@ -480,9 +477,7 @@ def test_clone_rejects_transcript_without_prompt_audio(monkeypatch, tmp_path, ca
with pytest.raises(SystemExit):
cli.main()
assert (
"--prompt-text/--prompt-file requires --prompt-audio" in capsys.readouterr().err
)
assert "--prompt-text/--prompt-file requires --prompt-audio" in capsys.readouterr().err
def test_batch_rejects_control_with_prompt_transcript(monkeypatch, tmp_path, capsys):
@@ -603,4 +598,4 @@ def test_batch_subcommand_passes_seed(monkeypatch, tmp_path):
assert len(dummy_model.calls) == 2
assert dummy_model.calls[0]["seed"] == 999
assert dummy_model.calls[1]["seed"] == 999
assert dummy_model.calls[1]["seed"] == 999
tests/test_validate.py
+2
View File
@@ -207,6 +207,7 @@ class TestValidateManifest:
audio = tmp_dir / "audio_8k.wav"
import numpy as np
samples = np.zeros(8000, dtype=np.float32)
sf.write(str(audio), samples, 8000)
@@ -240,6 +241,7 @@ class TestValidateManifest:
def test_cli_validate_exit_code(self, tmp_dir):
"""validate subcommand must exit 1 on validation error (missing audio)."""
import subprocess
manifest = tmp_dir / "bad.jsonl"
_write_manifest(manifest, [{"text": "hi", "audio": "/nonexistent/x.wav"}])