Skip to content
View mouse2p's full-sized avatar
🌴
On vacation
🌴
On vacation
  • nothing tech
  • Hong Kong
  • 16:43 (UTC +08:00)

Block or report mouse2p

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
mouse2p/README.md

Wang Yuanyi — Speech Recognition & End-to-End Spoken

EchoScribe EchoParley Python versions

关于我

做语音方向的研究与工程,专注两条线:

  • 语音识别(ASR)——流式解码、transducer/CTC 的搜索数学、码切换(英中混合)评估、延迟预算下的分块调度
  • 端到端语音对话(spoken dialogue)——离散语音单元、文本+语音统一 token 流、全双工轮次机制(barge-in、打断恢复)、会话级确定性重放

我相信语音工具的可复现性应该像编译器一样严格:默认全离线、种子即结果、每个测量数字都能在一台笔记本上重跑出来。

旗舰项目

流式语音识别工作台(Python · NumPy 核心 · torch CPU 可选)
  • RNN-T(transducer)与 CTC 解码:greedy、prefix-beam、标签同步分块解码,全部纯 NumPy 实现并带性质测试
  • 合成声学实验室:formant 合成器 + 四声调合成语言 + 植入转写,train→decode→score 全管线离线可复现
  • 码切换评估:WER(英)/ CER(中)/ 混合脚本评分规则,附对齐分解与 SVG 报告
  • 流式运行时:延迟预算、端点检测、热词偏置、会话录制重放
  • 600+ 增量提交 · 2400+ 测试 · CI 三版本矩阵
端到端语音对话框架(Python · NumPy 核心 · torch CPU 可选)
  • 语义语音单元(确定性 k-means / 可选 VQ)与统一交织 token 流(含轮次控制符)
  • 全双工轮次状态机:LISTEN → TAKE_FLOOR → SPEAK → YIELD → BARGE_IN → RECOVER,带延迟预算硬约束
  • 两方对话模拟器 + 植入轮次结构的合成对话实验室,指标对基准真值可证
  • 会话录制/重放字节级确定性;级联 vs 端到端同协议实测对比
  • 540+ 增量提交 · 2400+ 测试 · CI 三版本矩阵

方法与原则

  • 离线优先:默认测试与示例不需要网络、GPU、预训练权重或真实语音数据集
  • 合成数据 + 植入真值:所有报告数字来自带已知答案的合成实验,明确标注,绝不冒充真实基准成绩
  • 每提交验证:构建、测试、格式检查全绿才进历史;版本只走 patch 步进
  • 确定性即契约:同种子同输入 → 字节级一致输出,重放与直跑结果相同

构建旅程

EchoScribe 与 EchoParley 的模块旅程

贡献活动

contribution snake

echo = 回声 · scribe = 书写者 · parley = 会谈 —— 让机器听清,也让机器懂得轮流说话。

Popular repositories Loading

  1. echoparley echoparley Public

    End-to-end spoken-dialogue framework: deterministic semantic speech units, unified interleaved token streams with turn-control tokens, a duplex turn-taking state machine with barge-in and latency b…

    Python 23 236

  2. echoscribe echoscribe Public

    Streaming speech-recognition toolkit: transducer & CTC decoding in NumPy, latency-budgeted chunk streaming, hotword biasing, endpointing, code-switch English-Chinese evaluation (WER/CER/hybrid), an…

    Python

  3. mouse2p mouse2p Public

    Profile README