Support Custom OpenAI-compatible endpoints for speech-to-text transcription (Fixes #136)

addon/globalPlugins/AIHub/__init__.py

@@ -76,9 +76,11 @@ def getClient(self):
 			if not manager.isReady():
 				continue
 			api_key = manager.get_api_key()
-			base_url = manager.get_base_url() or BASE_URLs[manager.provider]
+			base_url = manager.get_base_url() or BASE_URLs.get(manager.provider)
+			if base_url:
+				base_url = base_url.rstrip("/")
 			if provider not in _OPTIONAL_API_KEY_PROVIDERS and (not api_key or not api_key.strip()):
-				return None
+				continue
 			organization = manager.get_api_key(use_org=True)
 			org_val = organization.split(":=", 1)[1] if organization and organization.count(":=") == 1 else None
 			return OpenAIClient(api_key=api_key, base_url=base_url, organization=org_val)

addon/globalPlugins/AIHub/apiclient/_client.py

@@ -58,7 +58,7 @@ class OpenAIClient:
 
 	def __init__(
 		self,
-		api_key: str,
+		api_key: Optional[str] = "",
 		base_url: str = "https://api.openai.com/v1",
 		organization: Optional[str] = None,
 	):
@@ -436,7 +436,8 @@ def _audio_text_create(
 		body, content_type = _build_audio_text_body(file, model, response_format, kwargs)
 		headers = _build_headers(self.api_key, self.organization)
 		headers["Content-Type"] = content_type
-		url = f"{self.base_url}{endpoint}"
+		base = (self.base_url or "").rstrip("/")
+		url = f"{base}{endpoint}"
 		req = urllib.request.Request(url, data=body, headers=headers, method="POST")
 		raw, _ct = _open_bytes(self._opener, req, timeout=120)
 		text = raw.decode("utf-8", errors="replace")
@@ -560,7 +561,10 @@ def configure_client_for_provider(
 	if clone and hasattr(client, "clone"):
 		client = client.clone()
 	manager = apikeymanager.get(provider)
-	client.base_url = manager.get_base_url(account_id=account_id) or BASE_URLs[provider]
+	base_url = manager.get_base_url(account_id=account_id) or BASE_URLs.get(provider)
+	if base_url:
+		base_url = base_url.rstrip("/")
+	client.base_url = base_url
 	api_key = manager.get_api_key(account_id=account_id)
 	# Ollama doesn't require an API key; supply a placeholder so the Bearer header is set
 	# (some Ollama-compatible reverse proxies reject empty Authorization headers).

addon/globalPlugins/AIHub/apikeymanager.py

@@ -324,7 +324,7 @@ def add_account(self, name, api_key, org_name="", org_key="", base_url="", set_a
 			account["base_url"] = _normalize_ollama_base_url((base_url or "").strip())
 		else:
 			account["base_url"] = None
-		if self.provider != Provider.Ollama and not account["api_key"]:
+		if self.provider not in _USER_ENDPOINT_PROVIDERS and not account["api_key"]:
 			raise ValueError("API key is required")
 		bucket.setdefault("accounts", []).append(account)
 		if set_active or not bucket.get("active_account_id"):

addon/globalPlugins/AIHub/configspec.py

@@ -43,6 +43,8 @@
 		"voxtralModel": f"string(default={VOXTRAL_MODELS[0]})",
 		"openaiTranscriptionAccountId": 'string(default="")',
 		"mistralTranscriptionAccountId": 'string(default="")',
+		"customOpenAITranscriptionAccountId": 'string(default="")',
+		"customOpenAITranscriptionModel": f"string(default={WHISPER_MODELS[0]})",
 		"whisper.cpp": {
 			"enabled": "boolean(default=False)",
 			"host": "string(default='http://127.0.0.1:8081')"

addon/globalPlugins/AIHub/consts.py

@@ -138,6 +138,7 @@ class TranscriptionProvider(StrEnum):
 	WHISPER_CPP = auto()
 	OPENAI = auto()
 	MISTRAL = auto()
+	CUSTOM_OPENAI = auto()
 
 
 TRANSCRIPTION_PROVIDERS = tuple(p.value for p in TranscriptionProvider)

addon/globalPlugins/AIHub/conversation_dialog.py

@@ -2694,6 +2694,18 @@ def _resolveDictationConfig(self):
 					and ("voxtral" in model_id.lower() or "transcribe" in model_id.lower())
 				):
 					transcription_model = model_id
+			elif transcription_provider == TranscriptionProvider.CUSTOM_OPENAI:
+				transcription_account_id = self.conf["audio"].get("customOpenAITranscriptionAccountId", "").strip()
+				if not transcription_account_id and account and account.get("provider") == Provider.CustomOpenAI:
+					transcription_account_id = account.get("id")
+				model_id = getattr(model, "id", "")
+				if (
+					model
+					and getattr(model, "provider", "") == Provider.CustomOpenAI
+					and isinstance(model_id, str)
+					and (model_id == "whisper-1" or "transcribe" in model_id.lower() or "whisper" in model_id.lower())
+				):
+					transcription_model = model_id
 		return {
 			"use_direct": use_direct,
 			"transcription_provider": transcription_provider,

addon/globalPlugins/AIHub/conversations.py

@@ -51,6 +51,7 @@ def _generate_next_value_(name, start, count, last_values):
 			"TOOL_MISTRAL_SPEECH_TO_TEXT": "tool/mistral/speech-to-text",
 			"TOOL_OPENAI_TTS": "tool/openai/tts",
 			"TOOL_OPENAI_TRANSCRIPTION": "tool/openai/transcription",
+			"TOOL_CUSTOM_OPENAI_TRANSCRIPTION": "tool/customopenai/transcription",
 		}
 		return mapping[name]
 
@@ -61,6 +62,7 @@ def _generate_next_value_(name, start, count, last_values):
 	TOOL_MISTRAL_SPEECH_TO_TEXT = auto()
 	TOOL_OPENAI_TTS = auto()
 	TOOL_OPENAI_TRANSCRIPTION = auto()
+	TOOL_CUSTOM_OPENAI_TRANSCRIPTION = auto()
 
 
 def normalize_conversation_format(value) -> ConversationFormat:

addon/globalPlugins/AIHub/conversations_manager_dialog.py

@@ -20,7 +20,7 @@
 from .tool_lyria_dialog import Lyria3ProToolDialog
 from .tool_mistral_ocr_dialog import MistralOCRToolDialog
 from .tool_mistral_transcription_dialog import MistralSpeechToTextToolDialog
-from .tool_openai_transcription_dialog import OpenAITranscriptionToolDialog
+from .tool_openai_transcription_dialog import OpenAITranscriptionToolDialog, CustomOpenAITranscriptionToolDialog
 from .tool_openai_tts_dialog import OpenAITTSToolDialog
 from .toolsmenu import open_tool_dialog_by_class
 from .tool_voxtral_tts_dialog import VoxtralTTSToolDialog
@@ -768,6 +768,7 @@ def _open_selected_entries(self, selected_entries):
 			ConversationFormat.TOOL_GOOGLE_LYRIA_PRO: Lyria3ProToolDialog,
 			ConversationFormat.TOOL_OPENAI_TTS: OpenAITTSToolDialog,
 			ConversationFormat.TOOL_OPENAI_TRANSCRIPTION: OpenAITranscriptionToolDialog,
+			ConversationFormat.TOOL_CUSTOM_OPENAI_TRANSCRIPTION: CustomOpenAITranscriptionToolDialog,
 		}
 		for data in loaded:
 			conv_format = normalize_conversation_format(data.get("format", ConversationFormat.GENERIC.value))

addon/globalPlugins/AIHub/plugin_mixins.py

@@ -324,9 +324,21 @@ def script_askQuestion(self, gesture):
 		from .model import getModels
 		from .recordthread import RecordThread
 		from . import conversation_dialog
+		from .transcription import get_transcription_provider
+		from .consts import TranscriptionProvider
+		from .apiclient import OpenAIClient, configure_client_for_provider
 
-		if not self.getClient():
-			return ui.message(NO_AUTHENTICATION_KEY_PROVIDED_MSG)
+		client = self.getClient()
+		if not client:
+			provider = get_transcription_provider(conf.get("audio", {}))
+			if provider == TranscriptionProvider.WHISPER_CPP.value:
+				client = object()
+			elif provider == TranscriptionProvider.CUSTOM_OPENAI.value and apikeymanager.get(Provider.CustomOpenAI).isReady():
+				client = configure_client_for_provider(OpenAIClient(), Provider.CustomOpenAI)
+			elif provider == TranscriptionProvider.MISTRAL.value and apikeymanager.get(Provider.MistralAI).isReady():
+				client = object()
+			else:
+				return ui.message(NO_AUTHENTICATION_KEY_PROVIDED_MSG)
 		if self._askAudioPlaying:
 			mci_stop_ask_audio()
 			self._askAudioPlaying = False
@@ -393,9 +405,21 @@ def script_askQuestion(self, gesture):
 	def script_toggleRecording(self, gesture):
 		from . import conversation_dialog
 		from .recordthread import RecordThread
+		from .transcription import get_transcription_provider
+		from .consts import TranscriptionProvider
+		from .apiclient import OpenAIClient, configure_client_for_provider
 
-		if not self.getClient():
-			return ui.message(NO_AUTHENTICATION_KEY_PROVIDED_MSG)
+		client = self.getClient()
+		if not client:
+			provider = get_transcription_provider(conf.get("audio", {}))
+			if provider == TranscriptionProvider.WHISPER_CPP.value:
+				client = object()
+			elif provider == TranscriptionProvider.CUSTOM_OPENAI.value and apikeymanager.get(Provider.CustomOpenAI).isReady():
+				client = configure_client_for_provider(OpenAIClient(), Provider.CustomOpenAI)
+			elif provider == TranscriptionProvider.MISTRAL.value and apikeymanager.get(Provider.MistralAI).isReady():
+				client = object()
+			else:
+				return ui.message(NO_AUTHENTICATION_KEY_PROVIDED_MSG)
 		dlg = conversation_dialog.activeChatDlg
 		if dlg and dlg.IsShown() and hasattr(dlg, "_prompt_ctrl_has_focus") and dlg._prompt_ctrl_has_focus():
 			dlg.onRecord(None)
@@ -404,5 +428,5 @@ def script_toggleRecording(self, gesture):
 			stop_worker_thread(self.recordThread)
 			self.recordThread = None
 			return
-		self.recordThread = RecordThread(self.getClient(), conf=conf["audio"])
+		self.recordThread = RecordThread(client, conf=conf["audio"])
 		self.recordThread.start()

addon/globalPlugins/AIHub/recordthread.py

@@ -24,6 +24,7 @@
 
 from . import apikeymanager
 from .apiclient import (
+	OpenAIClient,
 	Transcription,
 	transcribe_audio_mistral,
 	configure_client_for_provider,
@@ -57,16 +58,19 @@ def transcribe_audio_file(path, conf, client=None):
 		return None
 	provider = get_transcription_provider(conf)
 	if provider == TranscriptionProvider.WHISPER_CPP:
-		rt = RecordThread(client or object(), conf=conf)
+		rt = RecordThread(client or object(), conf=conf, transcriptionProvider=provider)
 		result = rt._transcribe_whisper_cpp(path)
 	elif provider == TranscriptionProvider.MISTRAL:
-		rt = RecordThread(client or object(), conf=conf)
+		rt = RecordThread(client or object(), conf=conf, transcriptionProvider=provider)
 		result = rt._transcribe_mistral(path)
+	elif provider == TranscriptionProvider.CUSTOM_OPENAI:
+		rt = RecordThread(client or OpenAIClient(), conf=conf, transcriptionProvider=provider)
+		result = rt._transcribe_custom_openai(path)
 	else:
 		if not client:
 			return None
 		client = configure_client_for_provider(client, Provider.OpenAI, clone=True)
-		rt = RecordThread(client, conf=conf)
+		rt = RecordThread(client, conf=conf, transcriptionProvider=provider)
 		result = rt._transcribe_openai(path)
 	if result is None:
 		return None
@@ -154,6 +158,15 @@ def __init__(
 				self._transcriptionAccountId = apikeymanager.get(Provider.MistralAI).get_active_account_id()
 			if not self._transcriptionModel:
 				self._transcriptionModel = (conf or {}).get("voxtralModel", "voxtral-mini-latest")
+		elif self._transcriptionProvider == TranscriptionProvider.CUSTOM_OPENAI:
+			if not self._transcriptionAccountId:
+				self._transcriptionAccountId = (conf or {}).get("customOpenAITranscriptionAccountId", "")
+			if self._transcriptionAccountId and not apikeymanager.get(Provider.CustomOpenAI).isReady(account_id=self._transcriptionAccountId):
+				self._transcriptionAccountId = ""
+			if not self._transcriptionAccountId:
+				self._transcriptionAccountId = apikeymanager.get(Provider.CustomOpenAI).get_active_account_id()
+			if not self._transcriptionModel:
+				self._transcriptionModel = (conf or {}).get("customOpenAITranscriptionModel", "") or (conf or {}).get("whisperModel", "whisper-1")
 
 	def run(self):
 		if self.audioFile:
@@ -424,6 +437,29 @@ def _transcribe_openai(self, filename):
 				response_format=self.responseFormat,
 			)
 
+	def _transcribe_custom_openai(self, filename):
+		"""Transcribe via Custom OpenAI-compatible endpoint."""
+		manager = apikeymanager.get(Provider.CustomOpenAI)
+		account_id = self._transcriptionAccountId or manager.get_active_account_id()
+		base_url = manager.get_base_url(account_id=account_id)
+		if not base_url or not base_url.strip():
+			# Translators: Text in recording thread status and error messages.
+			raise ValueError(_("No Custom OpenAI endpoint URL configured. Please add one in AI-Hub settings."))
+		client = self.client if (self.client and hasattr(self.client, "clone")) else OpenAIClient()
+		client = configure_client_for_provider(
+			client,
+			Provider.CustomOpenAI,
+			account_id=account_id,
+			clone=True,
+		)
+		model = self._transcriptionModel or (self.conf or {}).get("customOpenAITranscriptionModel", "") or (self.conf or {}).get("whisperModel", "whisper-1")
+		with open(filename, "rb") as audio_file:
+			return client.audio.transcriptions.create(
+				model=model,
+				file=audio_file,
+				response_format=self.responseFormat,
+			)
+
 	def process_transcription(self, filename):
 		if self._wantAbort:
 			return
@@ -434,6 +470,8 @@ def process_transcription(self, filename):
 				transcription = self._transcribe_whisper_cpp(filename)
 			elif provider == TranscriptionProvider.MISTRAL:
 				transcription = self._transcribe_mistral(filename)
+			elif provider == TranscriptionProvider.CUSTOM_OPENAI:
+				transcription = self._transcribe_custom_openai(filename)
 			else:
 				transcription = self._transcribe_openai(filename)
 		except Exception as err:

addon/globalPlugins/AIHub/settings_dialog.py

@@ -178,6 +178,8 @@ def makeSettings(self, settingsSizer):
 			_("OpenAI Whisper"),
 			# Translators: NVDA Preferences — AI-Hub category: Recording — Transcription provider list — Mistral Voxtral option.
 			_("Mistral Voxtral"),
+			# Translators: NVDA Preferences — AI-Hub category: Recording — Transcription provider list — Custom OpenAI option.
+			_("Custom OpenAI"),
 		]
 		self.transcriptionProviderChoice = providerGroup.addLabeledControl(
 			# Translators: NVDA Preferences — AI-Hub category: Recording — Provider — which engine handles dictation transcription.
@@ -224,6 +226,12 @@ def makeSettings(self, settingsSizer):
 			wx.Choice,
 			choices=[],
 		)
+		self.customOpenAITranscriptionAccountChoice = accountsMapGroup.addLabeledControl(
+			# Translators: NVDA Preferences — AI-Hub category: Recording — Provider account mapping — Custom OpenAI account drop-down label.
+			_("Custom OpenAI transcription account:"),
+			wx.Choice,
+			choices=[],
+		)
 		self._refreshTranscriptionAccountChoices()
 		recordingGroup.addItem(accountsMapSizer)
 
@@ -285,6 +293,7 @@ def _refreshTranscriptionAccountChoices(self):
 		audio_conf = conf.get("audio", {})
 		openai_id = audio_conf.get("openaiTranscriptionAccountId", "")
 		mistral_id = audio_conf.get("mistralTranscriptionAccountId", "")
+		custom_openai_id = audio_conf.get("customOpenAITranscriptionAccountId", "")
 		labels, ids, selected_idx = self._buildTranscriptionAccountChoices(Provider.OpenAI, openai_id)
 		self._openaiTranscriptionAccountIds = ids
 		self.openaiTranscriptionAccountChoice.SetItems(labels)
@@ -293,6 +302,11 @@ def _refreshTranscriptionAccountChoices(self):
 		self._mistralTranscriptionAccountIds = ids
 		self.mistralTranscriptionAccountChoice.SetItems(labels)
 		self.mistralTranscriptionAccountChoice.SetSelection(selected_idx)
+		if hasattr(self, "customOpenAITranscriptionAccountChoice"):
+			labels, ids, selected_idx = self._buildTranscriptionAccountChoices(Provider.CustomOpenAI, custom_openai_id)
+			self._customOpenAITranscriptionAccountIds = ids
+			self.customOpenAITranscriptionAccountChoice.SetItems(labels)
+			self.customOpenAITranscriptionAccountChoice.SetSelection(selected_idx)
 
 	def onManageAccounts(self, evt):
 		from .accounts_dialog import show_accounts_management
@@ -306,12 +320,12 @@ def onResize(self, evt):
 
 	def onTranscriptionProviderChange(self, evt):
 		idx = self.transcriptionProviderChoice.GetSelection()
-		is_whisper_cpp = idx == 0
-		is_openai = idx == 1
-		is_mistral = idx == 2
-		self.whisperHost.Enable(is_whisper_cpp)
-		self.openaiTranscriptionAccountChoice.Enable(is_openai)
-		self.mistralTranscriptionAccountChoice.Enable(is_mistral)
+		provider = TRANSCRIPTION_PROVIDERS[idx] if 0 <= idx < len(TRANSCRIPTION_PROVIDERS) else TranscriptionProvider.OPENAI.value
+		self.whisperHost.Enable(provider == TranscriptionProvider.WHISPER_CPP.value)
+		self.openaiTranscriptionAccountChoice.Enable(provider == TranscriptionProvider.OPENAI.value)
+		self.mistralTranscriptionAccountChoice.Enable(provider == TranscriptionProvider.MISTRAL.value)
+		if hasattr(self, "customOpenAITranscriptionAccountChoice"):
+			self.customOpenAITranscriptionAccountChoice.Enable(provider == TranscriptionProvider.CUSTOM_OPENAI.value)
 
 	def onTrimSilenceChange(self, evt):
 		self.minSilenceSec.Enable(self.trimSilenceCheckbox.GetValue())
@@ -359,6 +373,13 @@ def onSave(self):
 			if 0 <= mistral_idx < len(getattr(self, "_mistralTranscriptionAccountIds", []))
 			else ""
 		)
+		if hasattr(self, "customOpenAITranscriptionAccountChoice"):
+			custom_idx = self.customOpenAITranscriptionAccountChoice.GetSelection()
+			conf["audio"]["customOpenAITranscriptionAccountId"] = (
+				self._customOpenAITranscriptionAccountIds[custom_idx]
+				if 0 <= custom_idx < len(getattr(self, "_customOpenAITranscriptionAccountIds", []))
+				else ""
+			)
 		conf["audio"]["trimSilence"] = self.trimSilenceCheckbox.GetValue()
 		conf["audio"]["minSilenceSec"] = int(self.minSilenceSec.GetValue())
 		for key, item in self.chatFeedback.items():