From c17bd1e614b04f0b116e5c5d2f8a7b6870a53732 Mon Sep 17 00:00:00 2001 From: "paseo-bot[bot]" <266920839+paseo-bot[bot]@users.noreply.github.com> Date: Fri, 6 Nov 2026 03:20:13 +0000 Subject: [PATCH] docs(voice): scope prompt hints and translate the new section Vocabulary hints apply to compatible transcription endpoints; local speech providers ignore them. Keep the new section available at the same anchor in both documentation languages. --- public-docs/voice.md | 6 +++--- public-docs/voice.zh-CN.md | 11 +++++++++++ 2 files changed, 14 insertions(+), 3 deletions(-) diff --git a/public-docs/voice.md b/public-docs/voice.md index 214044556f..31394dad6f 100644 --- a/public-docs/voice.md +++ b/public-docs/voice.md @@ -142,14 +142,14 @@ Keys and base URLs follow the fallback rule above. - `PASEO_VOICE_LOCAL_STT_MODEL`, `PASEO_VOICE_LOCAL_TTS_MODEL`, local voice STT/TTS model IDs - `PASEO_DICTATION_LANGUAGE`, dictation STT language - `PASEO_VOICE_LANGUAGE`, voice mode STT language; falls back to `PASEO_DICTATION_LANGUAGE` when unset -- `PASEO_DICTATION_TRANSCRIPTION_PROMPT`, optional dictation transcription prompt. Unset by default; empty or whitespace-only is treated as unset +- `PASEO_DICTATION_TRANSCRIPTION_PROMPT`, optional OpenAI dictation transcription prompt (including compatible servers). Unset by default; empty or whitespace-only is treated as unset - `PASEO_VOICE_LOCAL_TTS_SPEAKER_ID`, `PASEO_VOICE_LOCAL_TTS_SPEED`, optional local voice TTS tuning ## Transcription Prompt -Paseo sends no transcription prompt by default. Whisper-family models use `prompt` as transcript context, so instruction text can distort the transcript or bias its language. Omitting it does not prevent every hallucination on silence or noise. +With the OpenAI provider, including compatible servers, Paseo sends no transcription prompt by default. Whisper-family models use `prompt` as transcript context, so instruction text can distort the transcript or bias its language. Omitting it does not prevent every hallucination on silence or noise. -Set `PASEO_DICTATION_TRANSCRIPTION_PROMPT` to send a prompt explicitly, for example to bias toward project vocabulary. An empty or whitespace-only value means no prompt. This affects dictation only; voice mode never sends a prompt. +For OpenAI dictation, set `PASEO_DICTATION_TRANSCRIPTION_PROMPT` to send a prompt explicitly, for example to bias toward project vocabulary. An empty or whitespace-only value means no prompt. Local providers ignore this setting, and voice mode never sends a prompt. ```bash PASEO_DICTATION_TRANSCRIPTION_PROMPT="Vocabulary: Paseo, TypeScript." diff --git a/public-docs/voice.zh-CN.md b/public-docs/voice.zh-CN.md index 02690e908e..fae45d1607 100644 --- a/public-docs/voice.zh-CN.md +++ b/public-docs/voice.zh-CN.md @@ -120,8 +120,19 @@ Paseo 在配置的 OpenAI base URL 下使用以下路径： - `PASEO_VOICE_LOCAL_STT_MODEL`、`PASEO_VOICE_LOCAL_TTS_MODEL`：本地语音 STT/TTS 模型 ID - `PASEO_DICTATION_LANGUAGE`：口述 STT 语言（仅限 OpenAI STT；本地 Parakeet 会忽略） - `PASEO_VOICE_LANGUAGE`：语音模式 STT 语言；未设置时退回到 `PASEO_DICTATION_LANGUAGE`（仅限 OpenAI STT；本地 Parakeet 会忽略） +- `PASEO_DICTATION_TRANSCRIPTION_PROMPT`：可选的 OpenAI 口述转录提示词（包括兼容服务器）。默认不设置；空值或仅含空白字符的值视为未设置 - `PASEO_VOICE_LOCAL_TTS_SPEAKER_ID`、`PASEO_VOICE_LOCAL_TTS_SPEED`：可选的本地语音 TTS 调优 +## 转录提示词 {#transcription-prompt} + +使用 OpenAI Provider（包括兼容服务器）进行口述时，Paseo 默认不发送转录提示词。Whisper 系列模型把 `prompt` 当作转录上下文，因此指令文本可能使转录失真或影响输出语言。不发送提示词也无法消除静音或噪声中的所有幻觉。 + +对于 OpenAI 口述，可以设置 `PASEO_DICTATION_TRANSCRIPTION_PROMPT` 来显式发送提示词，例如提示项目术语。空值或仅含空白字符的值表示不发送提示词。本地 Provider 会忽略此设置，语音模式也不会发送提示词。 + +```bash +PASEO_DICTATION_TRANSCRIPTION_PROMPT="Vocabulary: Paseo, TypeScript." +``` + ## 使用须知 {#operational-notes} 语音模式可以启动和控制 Agent。请像对待直接的 Agent 指令一样谨慎对待语音提示词，尤其是在指定工作目录或执行破坏性操作时。