Repository navigation
Releases: dat3133/audio2text-cli
Release list
audio2text 0.1.0
audio2text 0.1.0 is the first stable release for Windows/Linux x86_64.
Provides guided initialization, remembered storage locations, download progress, Qwen3-ASR/faster-whisper model management, offline file/batch transcription and JSON output for agents. Includes the Windows faster-whisper CUDA worker-exit fix from beta.6.
Download the platform executable or installer and verify SHA256SUMS. After npm publication, install with npm install -g audio2text-cli --registry=https://registry.npmjs.org, then run audio2text init for first-time setup. Upgrading an existing installation reuses its models, runtimes and configuration.
Installation and automated checks cover both platforms. Windows CUDA transcription has been checked on real hardware, and Linux transcription has been tested by the maintainer. These checks do not cover every model/GPU configuration. Natural-speech accuracy benchmarks remain future work; review mixed-language output. See the versioned README for usage and limitations.
audio2text 0.1.0-beta.6
修复
修复 Windows 使用 faster-whisper CUDA 时,音频已处理完、TXT 已写出,最后仍因 GPU 模型释放崩溃而返回 WORKER_FAILED 的问题(#1)。
CLI 现在在 TXT 和最终 JSON 完整写出后结束一次性推理子进程,由系统回收资源,绕开已知的 Windows CUDA 模型析构故障。推理错误、输出冲突、批量部分失败和用户取消仍保留原有状态与退出码。CPU、Linux 和 Qwen 的清理流程保持原样。
升级
npm install -g audio2text-cli@beta --registry=https://registry.npmjs.org
audio2text --version版本应显示 audio2text 0.1.0-beta.6。也可固定安装 audio2text-cli@0.1.0-beta.6。
升级只更新 CLI,原有模型、运行环境及初始化时保存的存储目录可以直接复用,无需重新下载或初始化。旧任务已生成的 TXT 会继续保留;重新转写时请使用新文件名,或明确选择 --overwrite。
如果此前在 PowerShell 使用了指向本地开发构建的临时 Set-Alias,请先重开终端,再检查 npm 安装版本。
验证与范围
- 新增 8 个子进程回归测试,覆盖结果和大 JSON 完整交付、自检、错误、取消、输出冲突、推理中原生失败及其他平台/后端的正常清理。
- Windows RTX 4060 Laptop GPU、whisper/base 和既有托管 CUDA 12.6 运行环境已实测自检、短语音 TXT/JSON、批量部分失败及错误处理;任务结束后显存恢复到测试前水平。
- 发布流程执行 Windows/Linux 测试、构建、安装、附件校验、匿名 GitHub 下载及 npm 注册表安装验证。
本版本仍为 Windows/Linux x86_64 预览版。完整 Linux 推理、所有 CUDA/模型组合及自然语音准确率基准仍待补齐;识别文本仍需复核。
附件及 SHA256SUMS 见本页;平台程序和安装说明见 本版本 README。模型与推理依赖由显式 audio2text init 准备,升级不自动下载它们。
audio2text 0.1.0-beta.5
audio2text 0.1.0-beta.5 预览版,支持 Windows/Linux x86_64。
下载进度
- 模型下载按当前文件显示进度条、百分比及已下载/总大小;续传从实际已有字节继续,服务器重新下载时同步重置。
- Python 和环境依赖下载启用 uv 原生进度,显示包名和下载字节。用户自己的
UV_NO_PROGRESS和 uv 调试设置继续生效。 - uv 安装包下载也显示进度;总大小未知时显示活动标记与已下载大小。
- 自有进度条按终端宽度缩短长文件名,只重绘当前行,正常刷新最多每秒 5 次。完成、失败和取消后结束进度行,随后分别进行校验、安装和自检。
- 已缓存文件跳过下载。stderr 非终端、
TERM=dumb或AUDIO2TEXT_PLAIN=1时使用普通日志;stdout 继续只输出最终 JSON,CLI 参数、JSON schema 和退出码不变。
升级
npm install -g audio2text-cli@beta --registry=https://registry.npmjs.org
audio2text --version
audio2text init也可下载对应平台程序或安装脚本,并核对 SHA256SUMS。安装只准备 CLI;模型和运行环境由显式 init 准备,初始化后可离线转写。
验证与限制
回归覆盖进度计算、中文和长文件名、窄终端、刷新限频、续传、中断、重试、校验失败、缓存复用、本地 HTTP 重定向及 JSON stdout 隔离。Windows 已用受控数据验证模型进度,并用固定 uv 0.11.32 和 36.56 KiB 小依赖验证真实原生下载条,无需下载 Torch 或模型权重。Linux CI 包含终端重绘、Ctrl+C 和原有向导回归。
当前仍是预览版。Linux 完整推理、默认托管 CUDA 配置、全部模型规格及自然录音准确率基准仍待补齐;请复核识别结果。对应版本 README。
audio2text 0.1.0-beta.4
audio2text 0.1.0-beta.4 预览版,支持 Windows/Linux x86_64。
本次更新
- 初始化向导的“准备计划”页新增“修改存储目录(模型和运行环境)”,推荐模式和高级模式均可使用。
- 支持中文、空格和相对路径;预填当前目录,Enter 保留,Ctrl+U 清空,Esc 返回,Ctrl+C 取消。页面显示模型、运行环境的实际路径,以及目标磁盘可用空间和预计下载量。
- 初始化及离线自检成功后记住所选位置,后续命令无需反复传入目录参数。位置优先级为
--data-dir、AUDIO2TEXT_HOME、已保存位置、平台默认目录。 - 默认目录通过很小的
location.json定位新目录;切换位置不搬迁或删除旧资源。取消、准备失败或保存位置失败保留原默认位置;普通--data-dir调用仍仅覆盖当前命令。 - 新目录的资源与磁盘检查独立进行,复用经校验的已有资源,避免沿用旧目录的安装状态。保留 Unicode 菜单重绘修复与 JSON stdout 接口。
安装或升级
npm install -g audio2text-cli@beta --registry=https://registry.npmjs.org
audio2text --version
audio2text init在最后的准备计划中选择“修改存储目录(模型和运行环境)”,例如输入 D:\audio2text,然后确认开始准备。该位置在自检成功后保存。
也可下载对应平台程序或安装脚本,并核对 SHA256SUMS。npm 及程序安装不携带模型和推理运行环境,由显式 init 准备。
验证与限制
包含存储位置优先级、配置提交及失败恢复测试,以及 Linux PTY 的目录编辑、返回、取消和菜单重绘测试。Windows 已使用现有本地 Whisper 模型完成真实自检和后续无目录参数的离线转写;原默认目录只生成位置记录。
当前仍是预览版。Linux 完整推理、默认托管 CUDA 配置、全部模型规格和自然录音准确率基准仍待补齐;请复核识别结果。使用说明见 对应版本 README。
audio2text 0.1.0-beta.3
audio2text 0.1.0-beta.3 is a preview release for Windows/Linux x86_64.
Fixes an initialization wizard bug where pressing Up/Down on long Chinese model descriptions could erase the heading and completed steps. The menu now wraps by terminal display width, scrolls within a bounded area, and uses numbered choices when the window is too small. Model details, back navigation and cancellation remain available.
Install or update from the official npm registry:
npm install -g audio2text-cli@beta --registry=https://registry.npmjs.org
audio2text --version
audio2text initModels and inference runtimes are prepared separately by init; updating the CLI does not require removing existing models. Release attachments include Windows/Linux executables and installers, the npm package, license notices and SHA256SUMS.
Full Linux inference, default managed CUDA configurations and natural-speech accuracy benchmarks remain under validation. See the versioned README for support status and limitations.
audio2text 0.1.0-beta.2
audio2text 0.1.0-beta.2 is a preview release for Windows/Linux x86_64.
Provides guided initialization, Qwen3-ASR/faster-whisper model management, local file/batch transcription and JSON output for agents. Models and inference runtimes are installed separately by audio2text init.
Download the platform executable or installer and verify SHA256SUMS. The npm package is published: install with npm install -g audio2text-cli@beta, then run audio2text init.
The npm registry assigned both beta and latest to this first published version and rejected removal of latest with HTTP 400. This accepted first-publication exception does not establish a stable release. Use @beta or @0.1.0-beta.2 explicitly. Future preview publication commands continue to specify beta.
This version fixes Windows installer handling of checksum manifests served as binary HTTP responses. Published executable and package payloads remain unchanged after publication.
Full Linux inference, managed CUDA configurations and natural-speech accuracy benchmarks remain under validation. See the versioned README for support status and limitations, and the repository README for current publication status.
audio2text 0.1.0-beta.1
audio2text 0.1.0-beta.1 is a preview release for Windows/Linux x86_64.
Provides guided initialization, Qwen3-ASR/faster-whisper model management, local file/batch transcription and JSON output for agents. Models and inference runtimes are installed separately by audio2text init.
Download the platform executable or installer and verify SHA256SUMS. After npm publication, install with npm install -g audio2text-cli@beta, then run audio2text init.
Full Linux inference, managed CUDA configurations and natural-speech accuracy benchmarks remain under validation. See the versioned README for support status and limitations.