feat: 升级语音识别模型并优化转写逻辑

- 语音识别模型从 whisper-1 升级到 gpt-4o-mini-transcribe - 根据用户语言设置自动添加简繁体中文提示词 - 录音转文字新增 dialog_id 参数，支持获取对话上下文提高识别准确率 - 移除前端语言手动选择功能，简化用户操作 - 添加参数空值保护 - 优化 reasoning_effort 参数逻辑，区分 gpt-5 和 gpt-5.1+ 版本
2026-03-03 16:02:08 +00:00 · 2026-01-05 02:26:36 +00:00 · 2026-01-05 02:26:36 +00:00 · 42a2eb56c7
commit 42a2eb56c7
parent 4b0f4e388c
5 changed files with 84 additions and 84 deletions
--- a/app/Http/Controllers/Api/DialogController.php
+++ b/app/Http/Controllers/Api/DialogController.php
@ -1306,11 +1306,7 @@ class DialogController extends AbstractController
     *
     * @apiParam {String} base64                语音base64
     * @apiParam {Number} duration              语音时长（毫秒）
-     * @apiParam {String} [language]            识别语言
-     * - 比如：zh
-     * - 默认：自动识别
-     * - 格式：符合 ISO_639 标准
-     * - 此参数不一定起效果，AI会根据语音和language参考翻译识别结果
+     * @apiParam {Number} [dialog_id]           会话ID，用于获取上下文提高识别准确率
     * @apiParam {String} [translate]           翻译识别结果
     * - 比如：zh
     * - 默认：不翻译结果
@ -1327,9 +1323,9 @@ class DialogController extends AbstractController
        //
        $path = "uploads/tmp/chat/" . date("Ym") . "/" . $user->userid . "/";
        $base64 = Request::input('base64');
-        $language = Request::input('language');
        $translate = Request::input('translate');
        $duration = intval(Request::input('duration'));
+        $dialogId = intval(Request::input('dialog_id'));
        if ($duration < 600) {
            return Base::retError('说话时间太短');
        }
@ -1342,17 +1338,35 @@ class DialogController extends AbstractController
            return Base::retError($data['msg']);
        }
        $recordData = $data['data'];
+        // 构建上下文提示词
+        $promptParts = [];
+        if ($user->lang === 'zh') {
+            $promptParts[] = "如果识别到中文，优先使用简体中文输出";
+        } elseif ($user->lang === 'zh-CHT') {
+            $promptParts[] = "如果識別到中文，優先使用繁體中文輸出";
+        }
+        // 获取最近的聊天上下文
+        if ($dialogId > 0) {
+            $contextTexts = WebSocketDialogMsg::whereDialogId($dialogId)
+                ->whereIn('type', ['text'])
+                ->orderByDesc('id')
+                ->limit(5)
+                ->get()
+                ->reverse()
+                ->map(fn($msg) => $msg->extractMessageContent(100))
+                ->filter()
+                ->values()
+                ->toArray();
+            if (!empty($contextTexts)) {
+                $promptParts[] = "对话上下文：" . implode("；", $contextTexts) . "。";
+            }
+        }
        // 转文字
        $extParams = [];
-        if ($language) {
-            $extParams = [
-                'language' => $language === 'zh-CHT' ? 'zh' : $language,
-                'prompt' => "将此语音识别为“" . Doo::getLanguages($language) . "”。",
-            ];
+        if (!empty($promptParts)) {
+            $extParams['prompt'] = implode("\n\n", $promptParts);
        }
-        $result = AI::transcriptions($recordData['file'], $extParams, [
-            'accept-language' => Request::header('Accept-Language', 'zh')
-        ]);
+        $result = AI::transcriptions($recordData['file'], $extParams);
        if (Base::isError($result)) {
            return $result;
        }
@ -1944,10 +1958,15 @@ class DialogController extends AbstractController
            return Base::retSuccess("success", $msg);
        }
        WebSocketDialog::checkDialog($msg->dialog_id);
+        // 根据用户语言构建提示词
+        $extParams = [];
+        if ($user->lang === 'zh') {
+            $extParams['prompt'] = "如果识别到中文，优先使用简体中文输出";
+        } elseif ($user->lang === 'zh-CHT') {
+            $extParams['prompt'] = "如果識別到中文，優先使用繁體中文輸出";
+        }
        //
-        $result = AI::transcriptions(public_path($msgData['path']), [], [
-            'accept-language' => Request::header('Accept-Language', 'zh')
-        ]);
+        $result = AI::transcriptions(public_path($msgData['path']), $extParams);
        if (Base::isError($result)) {
            return $result;
        }
--- a/app/Module/AI.php
+++ b/app/Module/AI.php
@ -271,6 +271,9 @@ class AI
    {
        Apps::isInstalledThrow('ai');

+        $extParams = $extParams ?: [];
+        $extHeaders = $extHeaders ?: [];
+
        if (!file_exists($filePath)) {
            return Base::retError("语音文件不存在");
        }
@ -287,7 +290,7 @@ class AI
        $result = Cache::remember($cacheKey, Carbon::now()->addDays(), function () use ($extParams, $extHeaders, $filePath, $audioProvider) {
            $post = array_merge($extParams, [
                'file' => new \CURLFile($filePath),
-                'model' => 'whisper-1',
+                'model' => 'gpt-4o-mini-transcribe',
            ]);
            $header = array_merge($extHeaders, [
                'Content-Type' => 'multipart/form-data',
@ -373,8 +376,9 @@ class AI
                    ]
                ],
            ];
-            if (self::shouldSendReasoningEffort($provider)) {
-                $payload['reasoning_effort'] = 'minimal';
+            $reasoningEffort = self::getReasoningEffort($provider);
+            if ($reasoningEffort !== null) {
+                $payload['reasoning_effort'] = $reasoningEffort;
            }
            $post = json_encode($payload);

@ -454,8 +458,9 @@ class AI
                    ]
                ],
            ];
-            if (self::shouldSendReasoningEffort($provider)) {
-                $payload['reasoning_effort'] = 'minimal';
+            $reasoningEffort = self::getReasoningEffort($provider);
+            if ($reasoningEffort !== null) {
+                $payload['reasoning_effort'] = $reasoningEffort;
            }
            $post = json_encode($payload);

@ -542,8 +547,9 @@ class AI
                    ]
                ],
            ];
-            if (self::shouldSendReasoningEffort($provider)) {
-                $payload['reasoning_effort'] = 'minimal';
+            $reasoningEffort = self::getReasoningEffort($provider);
+            if ($reasoningEffort !== null) {
+                $payload['reasoning_effort'] = $reasoningEffort;
            }
            $post = json_encode($payload);

@ -712,7 +718,7 @@ class AI

        return [
            'vendor' => 'openai',
-            'model' => 'whisper-1',
+            'model' => 'gpt-4o-mini-transcribe',
            'api_key' => $key,
            'base_url' => rtrim($baseUrl, '/'),
            'agency' => $agency,
@ -720,23 +726,37 @@ class AI
    }

    /**
-     * 是否需要附加 reasoning_effort 参数
+     * 获取 reasoning_effort 参数值
     * @param array $provider
-     * @return bool
+     * @return string|null 返回 'none'/'low' 或 null（不需要此参数）
     */
-    protected static function shouldSendReasoningEffort(array $provider): bool
+    protected static function getReasoningEffort(array $provider): ?string
    {
        if (($provider['vendor'] ?? '') !== 'openai') {
-            return false;
+            return null;
        }
        $model = $provider['model'] ?? '';
-        
-        // 匹配 gpt- 开头后跟数字的模型名称
-        if (preg_match('/^gpt-(\d+)/', $model, $matches)) {
-            return intval($matches[1]) >= 5;
+
+        // gpt-5.1 及之后版本支持 none
+        if (preg_match('/^gpt-(\d+)\.(\d+)/', $model, $matches)) {
+            $major = intval($matches[1]);
+            $minor = intval($matches[2]);
+            if ($major > 5 || ($major === 5 && $minor >= 1)) {
+                return 'none';
+            }
+            if ($major === 5) {
+                return 'low';
+            }
        }
-        
-        return false;
+
+        // gpt-5 (无小版本号) 使用 low
+        if (preg_match('/^gpt-(\d+)(?![.\d])/', $model, $matches)) {
+            if (intval($matches[1]) >= 5) {
+                return 'low';
+            }
+        }
+
+        return null;
    }

    /**
--- a/resources/assets/js/pages/manage/components/ChatInput/index.vue
+++ b/resources/assets/js/pages/manage/components/ChatInput/index.vue
@ -262,8 +262,7 @@
                    <div class="convert-body">
                        <div class="convert-content">
                            <div v-if="recordConvertSetting" class="convert-setting">
-                                <i class="taskfont" :class="{active: !!cacheTranscriptionLanguage}" @click="convertSetting('transcription', $event)">&#xe628;</i>
-                                <i class="taskfont" :class="{active: !!recordConvertTranslate}" @click="convertSetting('translate', $event)">&#xe795;</i>
+                                <i class="taskfont" :class="{active: !!recordConvertTranslate}" @click="convertSetting($event)">&#xe795;</i>
                            </div>
                            <div class="convert-input">
                                <Input
@ -599,7 +598,6 @@ export default {
            'cacheDialogs',
            'dialogMsgs',

-            'cacheTranscriptionLanguage',
            'cacheKeyboard',
            'keyboardShow',
            'keyboardHeight',
@ -1686,7 +1684,6 @@ export default {
                        dialog_id: this.dialogId,
                        base64: reader.result,
                        duration: this.recordDuration,
-                        language: this.cacheTranscriptionLanguage,
                        translate: this.recordConvertTranslate
                    },
                    method: 'post',
@ -1707,7 +1704,7 @@ export default {
            reader.readAsDataURL(this.recordBlob);
        },

-        async convertSetting(type, event) {
+        async convertSetting(event) {
            if (this.recordConvertStatus !== 1) {
                $A.messageWarning("请稍后再试...")
                return;
@ -1717,33 +1714,17 @@ export default {
                label: languageList[item],
                value: item
            }))
-            let active
-            if (type === 'transcription') {
-                // 语音转文字
-                list.unshift(...[
-                    {label: this.$L('选择识别语言'), value: '', disabled: true},
-                    {label: this.$L('自动识别'), value: '', divided: true},
-                ])
-                active = this.cacheTranscriptionLanguage
-            } else {
-                // 翻译
-                list.unshift(...[
-                    {label: this.$L('选择翻译结果'), value: '', disabled: true},
-                    {label: this.$L('不翻译结果'), value: '', divided: true},
-                ])
-                active = this.recordConvertTranslate
-            }
+            list.unshift(...[
+                {label: this.$L('选择翻译结果'), value: '', disabled: true},
+                {label: this.$L('不翻译结果'), value: '', divided: true},
+            ])
            this.$store.commit('menu/operation', {
                event,
                list,
-                active,
+                active: this.recordConvertTranslate,
                language: false,
-                onUpdate: async (language) => {
-                    if (type === 'transcription') {
-                        await this.$store.dispatch('setTranscriptionLanguage', language)
-                    } else {
-                        this.recordConvertTranslate = language
-                    }
+                onUpdate: (language) => {
+                    this.recordConvertTranslate = language
                    this.convertRecord()
                }
            })
--- a/resources/assets/js/store/actions.js
+++ b/resources/assets/js/store/actions.js
@ -1135,7 +1135,6 @@ export default {
                'cacheLoginEmail',
                'cacheFileSort',
                'cacheTranslationLanguage',
-                'cacheTranscriptionLanguage',
                'cacheTranslations',
                'cacheEmojis',
                'userInfo',
@ -1174,8 +1173,7 @@ export default {
                    'clientId',
                    'cacheServerUrl',
                    'cacheCalendarView',
-                    'cacheTranslationLanguage',
-                    'cacheTranscriptionLanguage'
+                    'cacheTranslationLanguage'
                ],
                array: [
                    'cacheUserBasic',
@ -1231,11 +1229,6 @@ export default {
                state.cacheTranslationLanguage = languageName;
            }

-            // TranscriptionLanguage 检查
-            if (typeof languageList[state.cacheTranscriptionLanguage] === "undefined") {
-                state.cacheTranscriptionLanguage = '';
-            }
-
            // 处理用户信息
            if (state.userInfo.userid) {
                state.userId = state.userInfo.userid = $A.runNum(state.userInfo.userid);
@ -4290,16 +4283,6 @@ export default {
        $A.IDBSave('cacheTranslationLanguage', language);
    },

-    /**
-     * 设置语音转文字语言
-     * @param state
-     * @param language
-     */
-    setTranscriptionLanguage({state}, language) {
-        state.cacheTranscriptionLanguage = language
-        $A.IDBSave('cacheTranscriptionLanguage', language);
-    },
-
    /** *****************************************************************************************/
    /** ************************************* loads *********************************************/
    /** *****************************************************************************************/
--- a/resources/assets/js/store/state.js
+++ b/resources/assets/js/store/state.js
@ -264,9 +264,6 @@ export default {
    cacheTranslationLanguage: '',
    cacheTranslations: [],

-    // 语音转文字（识别语言）
-    cacheTranscriptionLanguage: '',
-
    // 下拉菜单操作
    menuOperation: {},