Skip to content
Projects

fufuchiu/speechloom has 31 GitHub stars and 172 forks. Star changes are not measured yet. Last checked 2026-11-12 01:10 UTC. Last push 2026-10-01.

How these counts are measured

End-to-end speech model research: waveform encoding, causal decoding, codec tokens, joint losses and streaming.

Measured star growth

in 24h
Update delayed
in 7 days
Update delayed
in 15 days
Update delayed
in 1 month
Update delayed

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

README.md

Original

speechloom

端到端语音模型研究工具:音频编码、因果解码、联合训练与流式协议。

Chen Yuxuan

历史说明:本仓库由经过验证的补丁序列重建;2025 年至 2026 年 8 月的 Git 时间戳用于展示项目演进,不代表代码实际开发日期。

CI

快速开始

python -m venv .venv
source .venv/bin/activate
pip install -e '.[dev]'
python -m speechloom tokenize '大学' --audio-codes 0,128,255
python examples/codec_roundtrip.py
python examples/stream_text.py

文本采用无损 UTF-8 byte token。音频码本分配独立 ID 区间,避免模态碰撞。 训练时输出序列为 BOS + text + separator + audio_codes + EOS。

模型实验

pip install torch --index-url https://download.pytorch.org/whl/cpu
python examples/train_tiny.py

提供可以训练的紧凑研究原型和完整数据链路。示例使用合成输入,验证损失下降、 梯度和掩码正确性;不包含经过大规模训练的权重,也没有真实语料成绩声明。

文档

验证

python -m build
python -m pytest -q
ruff check .
ruff format --check .

MIT License · Chen Yuxuan

Explore related projects