ai_co_discuss 0.3.27 → 0.3.28

package/backend/server.py

@@ -667,7 +667,8 @@ def list_rooms(connection, params=None):
     identifier = submission_value(params["submission_id"]) if "submission_id" in params else None
     where = "WHERE EXISTS (SELECT 1 FROM room_creations WHERE room_id=rooms.id AND submission_id=?)" if identifier else ""
     rows = connection.execute(f"""SELECT rooms.*,
-        COALESCE(stats.message_count,0) AS message_count,COALESCE(stats.latest_id,0) AS latest_id
+        COALESCE(stats.message_count,0) AS message_count,COALESCE(stats.latest_id,0) AS latest_id,
+        (SELECT created_at FROM messages WHERE id=stats.latest_id) AS latest_at
         FROM rooms LEFT JOIN room_message_stats stats ON stats.room_id=rooms.id {where}
         GROUP BY rooms.id ORDER BY rooms.created_at, rooms.id""", (identifier,) if identifier else ())
     directory = {row["room_id"]: dict(row) for row in connection.execute("SELECT * FROM room_directory")}
@@ -1311,6 +1312,10 @@ def message_filters(room_id, params):
             raise ValueError("发言职责无效")
         clauses.append("COALESCE(details.kind,'opinion')=?")
         values.append(params["kind"])
+    if params.get("host") == "1":
+        # Host view: moderation notes plus whatever the current host said.
+        clauses.append("(COALESCE(details.kind,'opinion')='moderation' OR messages.author=(SELECT host FROM topic_settings WHERE room_id=?))")
+        values.append(room_id)
     if "submission_id" in params:
         clauses.append("submission_id=?")
         values.append(submission_value(params["submission_id"]))

package/docs/runner.md

@@ -90,9 +90,9 @@ API 成员可以担任专家或主持人。房间指定了项目目录时，它
 
 ## 朗读与语音输入（可选）
 
-在火山引擎控制台开通「语音合成2.0」（朗读）和「录音文件识别2.0」（语音输入），复制 API Key，填进网页“设置”的“语音”。朗读还可选音色、语速，改语气提示（默认是“像给朋友讲故事一样”的自然语气）。密钥只保存在本机，费用按你的账号计费。
+在火山引擎控制台开通「语音合成2.0」（朗读）和「录音文件识别2.0」（语音输入），复制 API Key，填进网页“设置”的“语音”。朗读可设一个通用音色，并按 AI 平台（Claude、Codex、Gemini 等）分别配音色，没配的平台、你自己的消息和程序消息用通用音色；还可选语速、改语气提示（默认是“像给朋友讲故事一样”的自然语气）。密钥只保存在本机，费用按你的账号计费。
 
-- 朗读：每条消息的工具栏有“朗读”（手机上在“…”菜单里），在消息里选中文字后点“听当前”从选中处开始；读完这条会接着读后面的消息直到最新，正在读的句子会高亮并自动滚到视野内，点一下高亮的句子暂停，再点继续；代码块跳过。
+- 朗读：每条消息的工具栏有“朗读”（手机上在“…”菜单里），在消息里选中文字后点“听当前”从选中处开始；读完这条会接着读后面的消息直到最新，点输入框右下角的播报按钮开启语音播报：之后的新回复会自动朗读回答摘要（没写摘要的短回答读全文，长回答读第一段），主持人的发言会报“主持人”；讨论中还会提示新一轮的发言顺序和轮到谁，执行失败、操作待你允许和讨论结束也会播一句。同时到达的回复依次播报，积压较多时只播最新 3 条。锁屏或切到其他 App 时继续播放，锁屏界面和耳机按键可以暂停、继续、跳到下一条；读过的段落会缓存在电脑上，重听不再计费。正在读的句子会高亮并自动滚到视野内，点一下高亮的句子暂停，再点继续；代码块跳过。
 - 语音输入：点输入框右下角的麦克风开始说话，再点一次转成文字插入输入框，单次最长 5 分钟。浏览器会先请求麦克风权限。
 
 两者都由本机服务调用火山引擎，手机等其他设备经已配对电脑使用，电脑关机时不可用。

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ai_co_discuss",
-  "version": "0.3.27",
+  "version": "0.3.28",
   "bin": {
     "codiscuss": "runner/codiscuss.mjs"
   },

package/runner/tts.mjs

@@ -1,6 +1,6 @@
 /** Read-aloud: synthesize short text with Volcengine Doubao TTS 2.0; the key stays in the local private store. */
-import { randomUUID } from 'node:crypto';
-import { existsSync, mkdirSync, readFileSync } from 'node:fs';
+import { createHash, randomUUID } from 'node:crypto';
+import { existsSync, mkdirSync, readdirSync, readFileSync, rmSync, statSync } from 'node:fs';
 import { join } from 'node:path';
 
 import { read_saved_api_key } from './discussion_api.mjs';
@@ -15,6 +15,14 @@ const TTS_TIMEOUT_MS = 45_000;
 export const TTS_MAX_TEXT = 300;
 const MAX_CONTEXT = 200;
 const SETTINGS_FILE = 'tts-settings.json';
+const SPEAKER_PATTERN = /^[\w.-]{1,128}$/;
+// Trailing silence the page may ask for at a paragraph or message end; it is part of the audio, so it also plays in the background.
+const MAX_SILENCE_MS = 3000;
+// Synthesized paragraphs are kept on this computer, so hearing the same text again is instant and not billed again.
+const CACHE_DIR = 'tts-cache';
+const CACHE_LIMIT = 400;
+const CACHE_PRUNE = 100;
+const MAX_MEMBER_VOICES = 50;
 // A default tone instruction (context_texts) so reading sounds told rather than recited; users may change or clear it.
 export const TTS_DEFAULT_CONTEXT = '请用自然放松的语气朗读，像给朋友讲故事一样，有轻重缓急，句子之间自然停顿。';
 export const TTS_DEFAULTS = { speaker: 'zh_female_vv_uranus_bigtts', speech_rate: 0, context: TTS_DEFAULT_CONTEXT };
@@ -55,20 +63,27 @@ const STATUS_MESSAGES = {
 export function tts_settings(home) {
   const path = join(home, SETTINGS_FILE);
   const saved = existsSync(path) ? JSON.parse(readFileSync(path, 'utf8')) : {};
-  return { ...TTS_DEFAULTS, ...saved, voices: TTS_VOICES, key_base_url: TTS_BASE_URL };
+  return { ...TTS_DEFAULTS, member_voices: {}, ...saved, voices: TTS_VOICES, key_base_url: TTS_BASE_URL };
 }
 
 /** Validate and save the read-aloud settings the page edits. */
 export function save_tts_settings(home, input) {
   const speaker = String(input.speaker ?? TTS_DEFAULTS.speaker).trim();
-  if (!/^[\w.-]{1,128}$/.test(speaker)) throw new Error('请填写有效的音色 ID');
+  if (!SPEAKER_PATTERN.test(speaker)) throw new Error('请填写有效的音色 ID');
   const speech_rate = Number(input.speech_rate ?? 0);
   if (!Number.isInteger(speech_rate) || speech_rate < -50 || speech_rate > 100) throw new Error('语速须为 -50 到 100 的整数');
   const context = String(input.context ?? '').trim();
   if (context.length > MAX_CONTEXT) throw new Error(`语气提示不超过 ${MAX_CONTEXT} 字`);
   const api_key_ref = input.api_key_ref || undefined;
   if (api_key_ref && !/^[a-f0-9-]{36}$/i.test(api_key_ref)) throw new Error('API Key 引用无效，请重新填写');
-  const settings = { speaker, speech_rate, context, ...(api_key_ref ? { api_key_ref } : {}) };
+  // Voices chosen per AI platform; members of other platforms, the user and program messages use the default voice.
+  const member_voices = {};
+  for (const [name, voice] of Object.entries(input.member_voices || {}).slice(0, MAX_MEMBER_VOICES)) {
+    if (!voice) continue;
+    if (String(name).length > 40 || !SPEAKER_PATTERN.test(voice)) throw new Error('成员音色无效，请重新选择');
+    member_voices[name] = voice;
+  }
+  const settings = { speaker, speech_rate, context, member_voices, ...(api_key_ref ? { api_key_ref } : {}) };
   mkdirSync(home, { recursive: true, mode: 0o700 });
   atomic_json(join(home, SETTINGS_FILE), settings);
   return tts_settings(home);
@@ -101,8 +116,24 @@ function word_marks(objects) {
   return objects.flatMap(item => item.sentence?.words || []).map(word => ({ w: String(word.word || ''), t: Math.round(Number(word.startTime || 0) * 1000) }));
 }
 
+/** Read a cached synthesis, refreshing its time so the most recently heard stay; null when absent or unreadable. */
+function cache_read(path) {
+  try { const value = JSON.parse(readFileSync(path, 'utf8')); atomic_json(path, value); return value; } catch { return null; }
+}
+
+/** Keep the cache bounded: past the limit, drop the least recently heard entries. */
+function cache_write(directory, path, value) {
+  mkdirSync(directory, { recursive: true, mode: 0o700 });
+  atomic_json(path, value);
+  const names = readdirSync(directory).filter(name => name.endsWith('.json'));
+  if (names.length <= CACHE_LIMIT) return;
+  const aged = names.map(name => ({ name, time: statSync(join(directory, name)).mtimeMs })).sort((a, b) => a.time - b.time);
+  for (const { name } of aged.slice(0, CACHE_PRUNE)) rmSync(join(directory, name), { force: true });
+}
+
 /**
  * Synthesize one paragraph and return it as base64 MP3 with word timings for highlighting.
+ * `speaker` may name another voice for this paragraph (a member's own voice); `silence_ms` appends trailing silence.
  * Requests of one reading pass share `section_id`, so the service keeps the earlier paragraphs as context.
  */
 export async function tts_speak(home, input, request = fetch) {
@@ -111,10 +142,17 @@ export async function tts_speak(home, input, request = fetch) {
   if (text.length > TTS_MAX_TEXT) throw new Error(`单次朗读不超过 ${TTS_MAX_TEXT} 字`);
   const settings = tts_settings(home);
   if (!settings.api_key_ref) throw new Error('尚未配置语音：请在设置的“语音”中填写火山引擎 API Key');
+  const speaker = SPEAKER_PATTERN.test(input.speaker || '') ? input.speaker : settings.speaker;
+  const silence = Math.max(0, Math.min(MAX_SILENCE_MS, Math.round(Number(input.silence_ms) || 0)));
+  const directory = join(home, CACHE_DIR);
+  const cache_path = join(directory, `${createHash('sha256').update(JSON.stringify([text, speaker, settings.speech_rate, settings.context, silence])).digest('hex')}.json`);
+  const cached = existsSync(cache_path) && cache_read(cache_path);
+  if (cached?.audio) return { ...cached, cached: true };
   const key = read_saved_api_key(home, TTS_BASE_URL, settings.api_key_ref);
   const section_id = /^[a-f0-9-]{36}$/i.test(input.section_id || '') ? input.section_id : undefined;
-  const additions = { ...(settings.context ? { context_texts: [settings.context] } : {}), ...(section_id ? { section_id } : {}) };
-  const body = { user: { uid: 'codiscuss' }, req_params: { text, speaker: settings.speaker,
+  const additions = { ...(settings.context ? { context_texts: [settings.context] } : {}), ...(section_id ? { section_id } : {}),
+    ...(silence ? { silence_duration: silence } : {}) };
+  const body = { user: { uid: 'codiscuss' }, req_params: { text, speaker,
     audio_params: { format: 'mp3', sample_rate: 24000, speech_rate: settings.speech_rate, enable_subtitle: true },
     ...(Object.keys(additions).length ? { additions: JSON.stringify(additions) } : {}) } };
   let response;
@@ -127,7 +165,11 @@ export async function tts_speak(home, input, request = fetch) {
   const objects = split_json_objects(await response.text());
   const chunks = [];
   for (const item of objects) {
-    if (item.code === TTS_DONE_CODE) return { audio: Buffer.concat(chunks).toString('base64'), format: 'mp3', marks: word_marks(objects) };
+    if (item.code === TTS_DONE_CODE) {
+      const result = { audio: Buffer.concat(chunks).toString('base64'), format: 'mp3', marks: word_marks(objects) };
+      cache_write(directory, cache_path, result);
+      return result;
+    }
     if (item.code !== 0) throw tts_error(item.code, item.message);
     if (item.data) chunks.push(Buffer.from(item.data, 'base64'));
   }