Skip to content
View FirmwareVersion's full-sized avatar
🎯
Focusing
🎯
Focusing
  • 武汉
  • 23:48 (UTC +08:00)

Block or report FirmwareVersion

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
FirmwareVersion/README.md

Zhang San — 语音识别 × 端到端语音大模型

speech recognition multi-speaker meetings offline-first

研究旅程:choravox

项目

项目 研究方向 已实现内容
choravox 多说话人会议语音 合成会议工厂(source-filter 语音合成 / image-method RIR / 麦克风阵列 / babble 噪声,全程植入真值)、能量与统计 VAD、UBM/PLDA 说话人嵌入与可选 torch 编码器、说话人日志(AHC / 谱聚类 / VB-HMM,重叠多标签输出)、帧级重叠检测、CTC/SOT 说话人归属 ASR 与目标说话人模式、强制对齐、cpWER/DER/JER 评测与条件切片、在线流式模式(增量聚类与延迟核算)、确定性 SVG/HTML 会议报告与完整 CLI

CI Release commits tests

520 次提交(2025.01 — 2026.09)· 41 个源码模块 · 4 个可运行离线示例 · 每次提交都经过构建、测试与格式检查。所有测量数字来自带植入真值的确定性合成数据:无真实语料、无预训练权重、全离线可复现。

技术栈

Python NumPy PyTorch pytest ruff GitHub Actions

活动

contribution snake

当前关注

  • 重叠语音的归属:SOT 序列化输出目标与说话人嵌入重 scoring 的联合解码。
  • 在线与离线的差距收敛:增量 leader-follower 聚类的出生/合并阈值,延迟-质量 Pareto 前沿核算。
  • 评测方法学:cpWER 匈牙利最优指派、DER 排序 collar 语义,以及合成切片上的诚实校准。

Popular repositories Loading

  1. choravox choravox Public

    Offline multi-speaker meeting-speech toolkit: planted-ground-truth synthetic meeting factory, VAD, speaker embeddings, diarization (AHC/spectral/VB-HMM), overlap detection, CTC speaker-attributed A…

    Python 22 191

  2. FirmwareVersion FirmwareVersion Public

    Profile README