From b4e0e461628c3e1cc08d92bc39501969bc071024 Mon Sep 17 00:00:00 2001
From: fengfeng-zi <xuf718@163.com>
Date: Sun, 3 May 2026 15:52:12 +0800
Subject: [PATCH] feat(stt): allow custom Azure speech transcription endpoint

---
 packages/components/src/speechToText.ts       | 21 ++++++++++--
 packages/components/src/utils.test.ts         | 32 +++++++++++++++++++
 .../ui-component/extended/SpeechToText.jsx    |  9 ++++++
 3 files changed, 60 insertions(+), 2 deletions(-)

diff --git a/packages/components/src/speechToText.ts b/packages/components/src/speechToText.ts
index e59368a5eae..c7964088101 100644
--- a/packages/components/src/speechToText.ts
+++ b/packages/components/src/speechToText.ts
@@ -14,6 +14,19 @@ const SpeechToTextType = {
     GROQ_WHISPER: 'groqWhisper'
 }
 
+export const buildAzureSpeechToTextUrl = (serviceRegion: string, apiVersion: string, baseUrl?: string) => {
+    const trimmedBaseUrl = baseUrl?.trim()
+    const base = trimmedBaseUrl
+        ? trimmedBaseUrl.replace(/\/+$/, '')
+        : `https://${serviceRegion}.cognitiveservices.azure.com/speechtotext/transcriptions:transcribe`
+
+    if (/[?&]api-version=/.test(base)) {
+        return base
+    }
+
+    return `${base}${base.includes('?') ? '&' : '?'}api-version=${encodeURIComponent(apiVersion)}`
+}
+
 export const convertSpeechToText = async (upload: IFileUpload, speechToTextConfig: ICommonObject, options: ICommonObject) => {
     if (speechToTextConfig) {
         const credentialId = speechToTextConfig.credentialId as string
@@ -76,8 +89,12 @@ export const convertSpeechToText = async (upload: IFileUpload, speechToTextConfi
             }
             case SpeechToTextType.AZURE_COGNITIVE: {
                 try {
-                    const baseUrl = `https://${credentialData.serviceRegion}.cognitiveservices.azure.com/speechtotext/transcriptions:transcribe`
                     const apiVersion = credentialData.apiVersion || '2024-05-15-preview'
+                    const azureSpeechToTextUrl = buildAzureSpeechToTextUrl(
+                        credentialData.serviceRegion,
+                        apiVersion,
+                        speechToTextConfig?.baseUrl
+                    )
 
                     const formData = new FormData()
                     const audioBlob = new Blob([new Uint8Array(audio_file)], { type: upload.type })
@@ -93,7 +110,7 @@ export const convertSpeechToText = async (upload: IFileUpload, speechToTextConfi
                     }
                     formData.append('definition', JSON.stringify(definition))
 
-                    const response = await axios.post(`${baseUrl}?api-version=${apiVersion}`, formData, {
+                    const response = await axios.post(azureSpeechToTextUrl, formData, {
                         headers: {
                             'Ocp-Apim-Subscription-Key': credentialData.azureSubscriptionKey,
                             Accept: 'application/json'
diff --git a/packages/components/src/utils.test.ts b/packages/components/src/utils.test.ts
index ba8238ee049..ae2633c779b 100644
--- a/packages/components/src/utils.test.ts
+++ b/packages/components/src/utils.test.ts
@@ -1,4 +1,5 @@
 import { removeInvalidImageMarkdown, convertRequireToImport, COMMONJS_REQUIRE_REGEX, IMPORT_EXTRACTION_REGEX } from './utils'
+import { buildAzureSpeechToTextUrl } from './speechToText'
 
 describe('removeInvalidImageMarkdown', () => {
     describe('strips non-http/https image markdown', () => {
@@ -229,3 +230,34 @@ describe('Import extraction regex (utils.ts line 1596 pattern)', () => {
         expect(extractModules('console.log("hello")')).toEqual([])
     })
 })
+
+describe('buildAzureSpeechToTextUrl', () => {
+    it('builds default regional URL', () => {
+        const url = buildAzureSpeechToTextUrl('eastus', '2024-05-15-preview')
+        expect(url).toBe(
+            'https://eastus.cognitiveservices.azure.com/speechtotext/transcriptions:transcribe?api-version=2024-05-15-preview'
+        )
+    })
+
+    it('uses custom baseUrl and appends api-version', () => {
+        const url = buildAzureSpeechToTextUrl(
+            'eastus',
+            '2024-05-15-preview',
+            'https://custom.example.com/speechtotext/transcriptions:transcribe/'
+        )
+        expect(url).toBe(
+            'https://custom.example.com/speechtotext/transcriptions:transcribe?api-version=2024-05-15-preview'
+        )
+    })
+
+    it('keeps existing api-version in custom baseUrl', () => {
+        const url = buildAzureSpeechToTextUrl(
+            'eastus',
+            '2024-05-15-preview',
+            'https://custom.example.com/speechtotext/transcriptions:transcribe?api-version=2023-10-01'
+        )
+        expect(url).toBe(
+            'https://custom.example.com/speechtotext/transcriptions:transcribe?api-version=2023-10-01'
+        )
+    })
+})
diff --git a/packages/ui/src/ui-component/extended/SpeechToText.jsx b/packages/ui/src/ui-component/extended/SpeechToText.jsx
index 2ca7fd95c28..313318db243 100644
--- a/packages/ui/src/ui-component/extended/SpeechToText.jsx
+++ b/packages/ui/src/ui-component/extended/SpeechToText.jsx
@@ -165,6 +165,15 @@ const speechToTextProviders = {
                 placeholder: 'en-US',
                 optional: true
             },
+            {
+                label: 'Base URL',
+                name: 'baseUrl',
+                type: 'string',
+                description:
+                    'Optional custom Azure Speech endpoint URL. Leave blank to use the default regional endpoint.',
+                placeholder: 'https://{region}.cognitiveservices.azure.com/speechtotext/transcriptions:transcribe',
+                optional: true
+            },
             {
                 label: 'Profanity Filter Mode',
                 name: 'profanityFilterMode',