Transcribe on OpenRouter too, not just OpenAI

OpenRouter mirrors OpenAI's /audio/transcriptions field for field, so one
multipart request serves both providers and only the key, the base URL and
the model id change. That puts a dozen speech models behind the key that was
already there for the cleanup, and makes a single OpenRouter key enough for
the whole chain.

Its transcription endpoint takes no hint field, so the words from Cleanup
rules are not sent there; they still reach the cleanup model as a glossary.
Timestamps switch to openai/whisper-1, the namespaced id of the only model
that returns segment times.

The API tab is now grouped by role rather than by service, because a key no
longer belongs to a single job: both keys sit at the top, the two jobs below.
Each provider keeps its own model, so switching back and forth does not
overwrite the other one's.

README.md

@@ -1,7 +1,7 @@
 # Dikte
 
-Press `Ctrl+Space`, talk, press again. The recording goes to OpenAI for
+Press `Ctrl+Space`, talk, press again. The recording goes to OpenAI or OpenRouter
-transcription, a model on OpenRouter cleans it up (dropping the *uh*s, the
+for transcription, a model on OpenRouter cleans it up (dropping the *uh*s, the
 restarts, the missing punctuation), and the result lands in your clipboard and
 is pasted into whatever window you were typing in.
 
@@ -32,12 +32,12 @@ dikte                        # the settings window opens on first run
 `install.sh` adds the `dikte` command, a menu entry, an autostart entry and the
 KDE shortcut.
 
-Two keys go in the settings window: **OpenAI** for speech to text
+Two keys go in the settings window: **OpenAI** and **OpenRouter**. Speech to text
-(`gpt-4o-transcribe`) and **OpenRouter** for the cleanup
+runs on either one (`gpt-4o-transcribe` by default), cleanup always on
-(`google/gemini-3.5-flash-lite` by default, any model on the list works). They
+OpenRouter (`google/gemini-3.5-flash-lite`), so a single OpenRouter key can
-fall back to `OPENAI_API_KEY` and `OPENROUTER_API_KEY`, and are stored in
+cover both. They fall back to `OPENAI_API_KEY` and `OPENROUTER_API_KEY`, and are
-`~/.config/dikte/config.json`, mode 600. Cleanup can be switched off, in which
+stored in `~/.config/dikte/config.json`, mode 600. Cleanup can be switched off,
-case the raw transcript is pasted.
+in which case the raw transcript is pasted.
 
 ## Using it
 
@@ -99,7 +99,7 @@ needs your user in the `input` group: `sudo usermod -aG input $USER`.
 ```
 dikte.py          entry point, tray icon, state machine, IPC
 audio.py          raw PCM capture through pw-record plus the level meter
-api.py            OpenAI transcription and OpenRouter cleanup (stdlib only)
+api.py            transcription on either provider, OpenRouter cleanup (stdlib only)
 worker.py         transcribe → clean up → clipboard → paste
 vad.py            deciding whether a recording holds speech at all
 filetranscribe.py file transcription: ffmpeg, chunking, timestamps

README.tr.md

@@ -1,8 +1,9 @@
 # Dikte
 
-`Ctrl+Space`'e bas, konuş, tekrar bas. Ses OpenAI'ye gidip yazıya çevrilir,
+`Ctrl+Space`'e bas, konuş, tekrar bas. Ses OpenAI'ye ya da OpenRouter'a gidip
-OpenRouter'daki bir model transkripti temizler (ıı'lar, tekrarlar, eksik
+yazıya çevrilir, OpenRouter'daki bir model transkripti temizler (ıı'lar,
-noktalama), sonuç panoya kopyalanır ve o an yazdığın pencereye yapıştırılır.
+tekrarlar, eksik noktalama), sonuç panoya kopyalanır ve o an yazdığın pencereye
+yapıştırılır.
 
 KDE Plasma 6 / Wayland için yazıldı. Sistem paketleri dışında bağımlılığı yok:
 sadece Python standart kütüphanesi ve PyQt6.
@@ -31,12 +32,13 @@ dikte                        # ilk açılışta ayarlar penceresi gelir
 `install.sh` `dikte` komutunu, menü girdisini, oturum açılışında otomatik
 başlatmayı ve KDE kısayolunu kurar.
 
-Ayarlar penceresinde iki anahtar istenir: sesi yazıya çeviren **OpenAI**
+Ayarlar penceresinde iki anahtar istenir: **OpenAI** ve **OpenRouter**. Sesi
-(`gpt-4o-transcribe`) ve transkripti temizleyen **OpenRouter** (varsayılan
+yazıya çevirme ikisinden birinde çalışır (varsayılan `gpt-4o-transcribe`),
-`google/gemini-3.5-flash-lite`, listedeki her model çalışır). Boş bırakırsan
+temizleme her zaman OpenRouter'da (`google/gemini-3.5-flash-lite`), yani tek bir
-`OPENAI_API_KEY` ve `OPENROUTER_API_KEY` kullanılır; anahtarlar
+OpenRouter anahtarı ikisine de yeter. Boş bırakırsan `OPENAI_API_KEY` ve
-`~/.config/dikte/config.json` içinde, izinler 600. Temizlemeyi tamamen
+`OPENROUTER_API_KEY` kullanılır; anahtarlar `~/.config/dikte/config.json`
-kapatabilirsin, o zaman ham transkript yapıştırılır.
+içinde, izinler 600. Temizlemeyi tamamen kapatabilirsin, o zaman ham transkript
+yapıştırılır.
 
 ## Kullanım
 
@@ -98,7 +100,7 @@ grubunda olmasını gerektirir: `sudo usermod -aG input $USER`.
 ```
 dikte.py          giriş noktası, tepsi simgesi, durum makinesi, IPC
 audio.py          pw-record ile ham PCM kaydı ve seviye ölçer
-api.py            OpenAI transkript + OpenRouter temizleme (yalnız stdlib)
+api.py            iki sağlayıcıda transkript + OpenRouter temizleme (yalnız stdlib)
 worker.py         transkript → temizleme → pano → yapıştırma
 vad.py            kayıtta gerçekten konuşma var mı kararı
 filetranscribe.py dosyadan transkript: ffmpeg, parçalama, zaman damgaları

api.py

@@ -1,5 +1,12 @@
-"""OpenAI (transcription) and OpenRouter (cleanup) calls, stdlib only."""
+"""OpenAI and OpenRouter calls, stdlib only.
 
+Transcription runs on either provider: OpenRouter mirrors OpenAI's
+/audio/transcriptions endpoint field for field, so one multipart request serves
+both and only the key, the base URL and the model id change. Cleanup is always
+OpenRouter.
+"""
+
+import collections
 import json
 import mimetypes
 import os
@@ -9,11 +16,20 @@ import urllib.request
 
 from i18n import t
 
-USER_AGENT = "dikte/1.0 (+https://github.com/yusufipk/dikte)"
+APP_URL = "https://github.com/yusufipk/dikte"
+USER_AGENT = f"dikte/1.0 (+{APP_URL})"
+OPENAI_URL = "https://api.openai.com/v1"
 OPENROUTER_URL = "https://openrouter.ai/api/v1"
 
-# Only whisper-1 returns segment-level timestamps.
+# Where a transcription request goes; built by config.Config.transcribe_target().
-TIMESTAMP_MODEL = "whisper-1"
+# `service` is the name the user sees in an error, `provider` the one the code
+# branches on.
+Target = collections.namedtuple("Target", "provider service api_key base_url model")
+
+
+def timestamp_model(provider):
+    """Only whisper-1 returns segment times, and OpenRouter namespaces the id."""
+    return "openai/whisper-1" if provider == "openrouter" else "whisper-1"
 
 
 class ApiError(Exception):
@@ -89,35 +105,44 @@ def _multipart(fields, file_field, file_path):
     return bytes(out), f"multipart/form-data; boundary={boundary}"
 
 
-def _transcribe_request(wav_path, api_key, model, language, prompt, base_url,
+def _headers(provider, api_key, content_type=None):
-                        response_format, granularity=None, timeout=300):
+    headers = {"Authorization": f"Bearer {api_key}", "User-Agent": USER_AGENT}
-    if not api_key:
+    if content_type:
-        raise ApiError(t("OpenAI API key is empty. Add it in Settings."))
+        headers["Content-Type"] = content_type
-    fields = [("model", model), ("response_format", response_format)]
+    if provider == "openrouter":
+        # What OpenRouter attributes the calls to on its app leaderboard.
+        headers["HTTP-Referer"] = APP_URL
+        headers["X-Title"] = "Dikte"
+    return headers
+
+
+def _transcribe_request(target, wav_path, language, prompt, response_format,
+                        granularity=None, timeout=300):
+    if not target.api_key:
+        raise ApiError(t("{service} API key is empty. Add it in Settings.",
+                         service=target.service))
+    fields = [("model", target.model), ("response_format", response_format)]
     if language and language != "auto":
         fields.append(("language", language))
-    if prompt:
+    # OpenRouter takes the hint field and throws it away, so spare it the bytes.
+    # The same words still reach the cleanup model as a glossary.
+    if prompt and target.provider == "openai":
         fields.append(("prompt", prompt))
     if granularity:
         fields.append(("timestamp_granularities[]", granularity))
     body, ctype = _multipart(fields, "file", wav_path)
-    headers = {
-        "Authorization": f"Bearer {api_key}",
-        "Content-Type": ctype,
-        "User-Agent": USER_AGENT,
-    }
     try:
         return _request(
-            f"{base_url.rstrip('/')}/audio/transcriptions", body, headers, timeout=timeout
+            f"{target.base_url.rstrip('/')}/audio/transcriptions", body,
+            _headers(target.provider, target.api_key, ctype), timeout=timeout,
         )
     except ApiError as exc:
-        raise explain(exc, "OpenAI") from None
+        raise explain(exc, target.service) from None
 
 
-def transcribe(wav_path, api_key, model="gpt-4o-transcribe", language="", prompt="",
+def transcribe(target, wav_path, language="", prompt="", timeout=300):
-               base_url="https://api.openai.com/v1", timeout=300):
     data = _transcribe_request(
-        wav_path, api_key, model, language, prompt, base_url, "json", timeout=timeout
+        target, wav_path, language, prompt, "json", timeout=timeout
     )
     text = (data.get("text") or "").strip()
     if not text:
@@ -125,12 +150,12 @@ def transcribe(wav_path, api_key, model="gpt-4o-transcribe", language="", prompt
     return text
 
 
-def transcribe_segments(wav_path, api_key, language="", prompt="",
+def transcribe_segments(target, wav_path, language="", prompt="", timeout=300):
-                        base_url="https://api.openai.com/v1", timeout=300):
     """[(start_seconds, text)] using whisper-1's verbose response."""
     data = _transcribe_request(
-        wav_path, api_key, TIMESTAMP_MODEL, language, prompt, base_url,
+        target._replace(model=timestamp_model(target.provider)),
-        "verbose_json", granularity="segment", timeout=timeout,
+        wav_path, language, prompt, "verbose_json",
+        granularity="segment", timeout=timeout,
     )
     segments = data.get("segments") or []
     out = []
@@ -148,7 +173,8 @@ def transcribe_segments(wav_path, api_key, language="", prompt="",
 
 def cleanup(text, api_key, model, system_prompt, base_url=OPENROUTER_URL, timeout=180):
     if not api_key:
-        raise ApiError(t("OpenRouter API key is empty. Add it in Settings."))
+        raise ApiError(t("{service} API key is empty. Add it in Settings.",
+                         service="OpenRouter"))
     payload = {
         "model": model,
         "temperature": 0,
@@ -157,18 +183,11 @@ def cleanup(text, api_key, model, system_prompt, base_url=OPENROUTER_URL, timeou
             {"role": "user", "content": f"<transcript>\n{text}\n</transcript>"},
         ],
     }
-    headers = {
-        "Authorization": f"Bearer {api_key}",
-        "Content-Type": "application/json",
-        "User-Agent": USER_AGENT,
-        "HTTP-Referer": "https://github.com/yusufipk/dikte",
-        "X-Title": "Dikte",
-    }
     try:
         data = _request(
             f"{base_url.rstrip('/')}/chat/completions",
             json.dumps(payload).encode("utf-8"),
-            headers,
+            _headers("openrouter", api_key, "application/json"),
             timeout=timeout,
         )
     except ApiError as exc:
@@ -199,7 +218,8 @@ def _get_json(url, headers, timeout=20):
 def openrouter_key_status(api_key):
     """Check the key against OpenRouter's own /key endpoint."""
     if not api_key:
-        raise ApiError(t("OpenRouter API key is empty. Add it in Settings."))
+        raise ApiError(t("{service} API key is empty. Add it in Settings.",
+                         service="OpenRouter"))
     try:
         data = _get_json(f"{OPENROUTER_URL}/key",
                          {"Authorization": f"Bearer {api_key}", "User-Agent": USER_AGENT})
@@ -213,18 +233,32 @@ def openrouter_key_status(api_key):
              usage=round(float(usage or 0), 3), limit=round(float(limit), 3))
 
 
-def openrouter_models(api_key=""):
+def openrouter_models(api_key="", transcription=False):
-    """Model ids available on OpenRouter (no key required)."""
+    """Model ids available on OpenRouter (no key required).
+
+    `transcription` narrows the list to the speech-to-text models, the only ones
+    /audio/transcriptions accepts. The filter is applied again on the result,
+    because a query parameter the API stops honouring would otherwise quietly
+    hand back all several hundred models.
+    """
+    url = f"{OPENROUTER_URL}/models"
+    if transcription:
+        url += "?output_modalities=transcription"
     headers = {"User-Agent": USER_AGENT}
     if api_key:
         headers["Authorization"] = f"Bearer {api_key}"
-    data = _get_json(f"{OPENROUTER_URL}/models", headers)
+    models = _get_json(url, headers).get("data", [])
-    return sorted(m["id"] for m in data.get("data", []) if m.get("id"))
+    if transcription:
+        models = [m for m in models
+                  if "transcription" in (m.get("architecture") or {}).get(
+                      "output_modalities", [])]
+    return sorted(m["id"] for m in models if m.get("id"))
 
 
-def openai_models(api_key, base_url="https://api.openai.com/v1"):
+def openai_models(api_key, base_url=OPENAI_URL):
     if not api_key:
-        raise ApiError(t("OpenAI API key is empty. Add it in Settings."))
+        raise ApiError(t("{service} API key is empty. Add it in Settings.",
+                         service="OpenAI"))
     try:
         data = _get_json(
             f"{base_url.rstrip('/')}/models",

config.py

@@ -5,6 +5,7 @@ import json
 import os
 import pathlib
 
+import api
 import i18n
 
 
@@ -92,7 +93,9 @@ DEFAULTS = {
     "openai_base_url": "https://api.openai.com/v1",
     "openrouter_api_key": "",
     "openrouter_base_url": "https://openrouter.ai/api/v1",
-    "transcribe_model": "gpt-4o-transcribe",
+    "transcribe_provider": "openai",  # openai | openrouter
+    "transcribe_model": "gpt-4o-transcribe",           # used when provider is openai
+    "openrouter_transcribe_model": "openai/gpt-4o-transcribe",
     "language": "tr",
     "transcribe_prompt": "",
     "cleanup_enabled": True,
@@ -184,6 +187,15 @@ class Config:
     def openrouter_key(self):
         return self["openrouter_api_key"].strip() or os.environ.get("OPENROUTER_API_KEY", "").strip()
 
+    def transcribe_target(self):
+        """Key, endpoint and model for whichever provider does speech to text."""
+        if self["transcribe_provider"] == "openrouter":
+            return api.Target("openrouter", "OpenRouter", self.openrouter_key(),
+                              self["openrouter_base_url"],
+                              self["openrouter_transcribe_model"])
+        return api.Target("openai", "OpenAI", self.openai_key(),
+                          self["openai_base_url"], self["transcribe_model"])
+
     def cleanup_prompt(self, with_timestamps=False):
         turkish = i18n.language() == "tr"
         prompt = self["cleanup_prompt"].strip() or default_cleanup_prompt()

dikte.py

@@ -320,7 +320,9 @@ def main():
     server.newConnection.connect(on_connection)
     app.aboutToQuit.connect(dikte.shutdown)
 
-    if command == "settings" or not dikte.conf.openai_key():
+    # No key for the chosen transcription provider means nothing can work yet,
+    # so the settings window is the only useful thing to open.
+    if command == "settings" or not dikte.conf.transcribe_target().api_key:
         dikte.open_settings()
     elif command == "toggle":
         QTimer.singleShot(0, dikte.toggle)

filetranscribe.py

@@ -74,6 +74,7 @@ class FileTranscriber(QObject):
             if len(chunks) > 1:
                 self.progress.emit(t("Splitting into {count} chunks…", count=len(chunks)))
 
+            target = conf.transcribe_target()
             pieces = []
             for index, (chunk_path, offset) in enumerate(chunks, start=1):
                 self._check()
@@ -82,11 +83,10 @@ class FileTranscriber(QObject):
                 )
                 if timestamps:
                     segments = api.transcribe_segments(
+                        target,
                         chunk_path,
-                        conf.openai_key(),
                         language=conf["language"],
                         prompt=conf["transcribe_prompt"],
-                        base_url=conf["openai_base_url"],
                     )
                     pieces.extend(
                         f"[{format_timestamp(start + offset)}] {text}"
@@ -94,12 +94,10 @@ class FileTranscriber(QObject):
                     )
                 else:
                     pieces.append(api.transcribe(
+                        target,
                         chunk_path,
-                        conf.openai_key(),
-                        model=conf["transcribe_model"],
                         language=conf["language"],
                         prompt=conf["transcribe_prompt"],
-                        base_url=conf["openai_base_url"],
                     ))
 
             text = "\n".join(pieces) if timestamps else " ".join(pieces)

i18n.py

@@ -93,10 +93,8 @@ TR = {
         "ydotool hatası: {error}\nydotoold çalışıyor mu? (systemctl --user status ydotool)",
 
     # --- api errors ----------------------------------------------------
-    "OpenAI API key is empty. Add it in Settings.":
+    "{service} API key is empty. Add it in Settings.":
-        "OpenAI API anahtarı boş. Ayarlar'dan gir.",
+        "{service} API anahtarı boş. Ayarlar'dan gir.",
-    "OpenRouter API key is empty. Add it in Settings.":
-        "OpenRouter API anahtarı boş. Ayarlar'dan gir.",
     "Transcript came back empty.": "Transkript boş döndü.",
     "The cleanup model returned an empty reply.": "Temizleme modeli boş yanıt döndü.",
     "Could not connect: {reason}": "Bağlantı kurulamadı: {reason}",
@@ -146,14 +144,23 @@ TR = {
         "Ses kayıtlarını sakla (~/.local/share/dikte/recordings)",
 
     # --- settings: api --------------------------------------------------
-    "OpenAI: speech to text": "OpenAI: sesi yazıya çevirme",
+    "Keys": "Anahtarlar",
-    "OpenRouter: transcript cleanup": "OpenRouter: transkripti temizleme",
+    "Speech to text": "Sesi yazıya çevirme",
+    "Transcript cleanup": "Transkripti temizleme",
     "API key": "API anahtarı",
     "Model": "Model",
+    "Provider": "Sağlayıcı",
     "sk-… (falls back to OPENAI_API_KEY)": "sk-… (boşsa OPENAI_API_KEY kullanılır)",
     "sk-or-… (falls back to OPENROUTER_API_KEY)": "sk-or-… (boşsa OPENROUTER_API_KEY kullanılır)",
-    "Test key": "Anahtarı test et",
+    "Test": "Test et",
     "Trying…": "Deneniyor…",
+    "Runs on OpenAI, with the key above.":
+        "Yukarıdaki anahtarla OpenAI üzerinde çalışır.",
+    "Runs on OpenRouter, with the key above. Word hints reach the cleanup "
+    "model only.":
+        "Yukarıdaki anahtarla OpenRouter üzerinde çalışır. Kelime ipuçları "
+        "yalnızca temizleme modeline gider.",
+    "Runs on OpenRouter.": "OpenRouter üzerinde çalışır.",
     "Connection works. {count} audio models visible.":
         "Bağlantı tamam. {count} ses modeli görünüyor.",
     "Clean the transcript with a model": "Transkripti bir modelle temizle",
@@ -187,10 +194,10 @@ TR = {
     "Audio and video files": "Ses ve video dosyaları",
     "All files": "Tüm dosyalar",
     "Add timestamps": "Zaman damgası ekle",
-    "Prefixes every segment with [mm:ss]. Uses whisper-1, the only model that "
+    "Prefixes every segment with [mm:ss]. Uses whisper-1 on whichever provider "
-    "returns segment times.":
+    "you picked, the only model that returns segment times.":
         "Her bölümün başına [dd:ss] koyar. Bölüm zamanı döndüren tek model olan "
-        "whisper-1 kullanılır.",
+        "whisper-1, seçtiğin sağlayıcı üzerinden kullanılır.",
     "Run the cleanup model afterwards": "Sonrasında temizleme modelinden geçir",
     "Transcribe": "Yazıya çevir",
     "Stop": "Durdur",

settings_ui.py

@@ -25,7 +25,18 @@ LANGUAGES = [
     ("German", "de"), ("French", "fr"), ("Spanish", "es"), ("Arabic", "ar"),
 ]
 CORNERS = ["bottom-left", "bottom-right", "top-left", "top-right"]
-TRANSCRIBE_MODELS = ["gpt-4o-transcribe", "gpt-4o-mini-transcribe", "whisper-1"]
+TRANSCRIBE_PROVIDERS = [("OpenAI", "openai"), ("OpenRouter", "openrouter")]
+# Starting points for the model box; "Fetch model list" replaces them with
+# whatever the provider offers today.
+TRANSCRIBE_MODELS = {
+    "openai": ["gpt-4o-transcribe", "gpt-4o-mini-transcribe", "whisper-1"],
+    "openrouter": [
+        "openai/gpt-4o-transcribe", "openai/gpt-4o-mini-transcribe",
+        "openai/whisper-1", "openai/whisper-large-v3",
+        "openai/whisper-large-v3-turbo", "mistralai/voxtral-mini-transcribe",
+        "deepgram/nova-3", "google/chirp-3",
+    ],
+}
 CLEANUP_MODELS = [
     "google/gemini-3.5-flash-lite", "google/gemini-3.1-flash-lite",
     "google/gemini-2.5-flash-lite", "anthropic/claude-haiku-4.5",
@@ -40,6 +51,7 @@ class SettingsWindow(QDialog):
     applied = pyqtSignal()
 
     _models_loaded = pyqtSignal(list, str)
+    _transcribe_models_loaded = pyqtSignal(list, str)
     _test_done = pyqtSignal(bool, str)
     _or_test_done = pyqtSignal(bool, str)
 
@@ -47,6 +59,10 @@ class SettingsWindow(QDialog):
         super().__init__(parent)
         self.conf = conf
         self.launch_command = launch_command
+        # Each provider keeps its own transcription model, so switching the
+        # provider back and forth never overwrites the other one's.
+        self._models = {"openai": "", "openrouter": ""}
+        self._shown_provider = ""
         self.transcriber = FileTranscriber(conf, self)
         self.setWindowTitle(t("Dikte Settings"))
         self.resize(680, 640)
@@ -71,6 +87,7 @@ class SettingsWindow(QDialog):
         layout.addWidget(buttons)
 
         self._models_loaded.connect(self._on_models_loaded)
+        self._transcribe_models_loaded.connect(self._on_transcribe_models_loaded)
         self._test_done.connect(self._on_test_done)
         self._or_test_done.connect(self._on_or_test_done)
         self.transcriber.progress.connect(self._on_file_progress)
@@ -156,56 +173,69 @@ class SettingsWindow(QDialog):
         page = QWidget()
         outer = QVBoxLayout(page)
 
-        oai = QGroupBox(t("OpenAI: speech to text"))
+        # Keys first, then the two jobs, because OpenRouter can now do both of
-        oai_form = QFormLayout(oai)
+        # them and a key no longer belongs to a single job.
+        keys = QGroupBox(t("Keys"))
+        keys_form = QFormLayout(keys)
         self.openai_key = QLineEdit()
         self.openai_key.setEchoMode(QLineEdit.EchoMode.Password)
         self.openai_key.setPlaceholderText(t("sk-… (falls back to OPENAI_API_KEY)"))
-        oai_form.addRow(t("API key"), self.openai_key)
+        self.test_button = QPushButton(t("Test"))
-
-        self.transcribe_model = QComboBox()
-        self.transcribe_model.setEditable(True)
-        self.transcribe_model.addItems(TRANSCRIBE_MODELS)
-        oai_form.addRow(t("Model"), self.transcribe_model)
-
-        self.test_button = QPushButton(t("Test key"))
         self.test_button.clicked.connect(self._test_openai)
         self.test_label = QLabel("")
         self.test_label.setWordWrap(True)
-        row = QHBoxLayout()
+        keys_form.addRow("OpenAI", self._row(self.openai_key, self.test_button))
-        row.addWidget(self.test_button)
+        keys_form.addRow("", self.test_label)
-        row.addWidget(self.test_label, 1)
-        oai_form.addRow("", self._wrap(row))
-        outer.addWidget(oai)
-
-        orr = QGroupBox(t("OpenRouter: transcript cleanup"))
-        orr_form = QFormLayout(orr)
-        self.cleanup_enabled = QCheckBox(t("Clean the transcript with a model"))
-        orr_form.addRow("", self.cleanup_enabled)
 
         self.openrouter_key = QLineEdit()
         self.openrouter_key.setEchoMode(QLineEdit.EchoMode.Password)
         self.openrouter_key.setPlaceholderText(t("sk-or-… (falls back to OPENROUTER_API_KEY)"))
-        orr_form.addRow(t("API key"), self.openrouter_key)
+        self.or_test_button = QPushButton(t("Test"))
+        self.or_test_button.clicked.connect(self._test_openrouter)
+        self.or_test_label = QLabel("")
+        self.or_test_label.setWordWrap(True)
+        keys_form.addRow("OpenRouter", self._row(self.openrouter_key, self.or_test_button))
+        keys_form.addRow("", self.or_test_label)
+        outer.addWidget(keys)
+
+        stt = QGroupBox(t("Speech to text"))
+        stt_form = QFormLayout(stt)
+        self.transcribe_provider = QComboBox()
+        for label, value in TRANSCRIBE_PROVIDERS:
+            self.transcribe_provider.addItem(label, value)
+        stt_form.addRow(t("Provider"), self.transcribe_provider)
+
+        self.transcribe_model = QComboBox()
+        self.transcribe_model.setEditable(True)
+        self.refresh_transcribe_models = QPushButton(t("Fetch model list"))
+        self.refresh_transcribe_models.clicked.connect(self._load_transcribe_models)
+        stt_form.addRow(t("Model"),
+                        self._row(self.transcribe_model, self.refresh_transcribe_models))
+        # Spanning rows: in the narrow field column a wrapped label gets a height
+        # that fits one line, and the rest of the text is cut off.
+        self.transcribe_note = QLabel("")
+        self.transcribe_note.setWordWrap(True)
+        stt_form.addRow(self.transcribe_note)
+        self.transcribe_status = QLabel("")
+        self.transcribe_status.setWordWrap(True)
+        stt_form.addRow(self.transcribe_status)
+        self.transcribe_provider.currentIndexChanged.connect(self._provider_changed)
+        outer.addWidget(stt)
+
+        orr = QGroupBox(t("Transcript cleanup"))
+        orr_form = QFormLayout(orr)
+        self.cleanup_enabled = QCheckBox(t("Clean the transcript with a model"))
+        orr_form.addRow("", self.cleanup_enabled)
 
         self.cleanup_model = QComboBox()
         self.cleanup_model.setEditable(True)
         self.cleanup_model.addItems(CLEANUP_MODELS)
         self.refresh_models = QPushButton(t("Fetch model list"))
         self.refresh_models.clicked.connect(self._load_models)
-        model_row = QHBoxLayout()
+        orr_form.addRow(t("Model"), self._row(self.cleanup_model, self.refresh_models))
-        model_row.addWidget(self.cleanup_model, 1)
+        self.models_label = QLabel(t("Runs on OpenRouter."))
-        model_row.addWidget(self.refresh_models)
-        orr_form.addRow(t("Model"), self._wrap(model_row))
-
-        self.or_test_button = QPushButton(t("Test key"))
-        self.or_test_button.clicked.connect(self._test_openrouter)
-        self.models_label = QLabel("")
         self.models_label.setWordWrap(True)
-        test_row = QHBoxLayout()
+        orr_form.addRow(self.models_label)
-        test_row.addWidget(self.or_test_button)
-        test_row.addWidget(self.models_label, 1)
-        orr_form.addRow("", self._wrap(test_row))
         outer.addWidget(orr)
         outer.addStretch(1)
         return page
@@ -255,8 +285,8 @@ class SettingsWindow(QDialog):
 
         self.file_timestamps = QCheckBox(t("Add timestamps"))
         self.file_timestamps.setToolTip(
-            t("Prefixes every segment with [mm:ss]. Uses whisper-1, the only model "
+            t("Prefixes every segment with [mm:ss]. Uses whisper-1 on whichever "
-              "that returns segment times.")
+              "provider you picked, the only model that returns segment times.")
         )
         layout.addWidget(self.file_timestamps)
 
@@ -385,10 +415,15 @@ class SettingsWindow(QDialog):
         return page
 
     @staticmethod
-    def _wrap(layout):
+    def _row(*widgets):
-        widget = QWidget()
+        """Widgets side by side in one form row; the first one takes the space."""
-        widget.setLayout(layout)
+        layout = QHBoxLayout()
-        return widget
+        layout.setContentsMargins(0, 0, 0, 0)
+        for index, widget in enumerate(widgets):
+            layout.addWidget(widget, 1 if index == 0 else 0)
+        holder = QWidget()
+        holder.setLayout(layout)
+        return holder
 
     # ---- load / save ----------------------------------------------------
 
@@ -408,9 +443,13 @@ class SettingsWindow(QDialog):
         self.keep_audio.setChecked(conf["keep_audio"])
 
         self.openai_key.setText(conf["openai_api_key"])
-        self.transcribe_model.setCurrentText(conf["transcribe_model"])
-        self.cleanup_enabled.setChecked(conf["cleanup_enabled"])
         self.openrouter_key.setText(conf["openrouter_api_key"])
+        self._models = {"openai": conf["transcribe_model"],
+                        "openrouter": conf["openrouter_transcribe_model"]}
+        self._shown_provider = ""
+        self._select_data(self.transcribe_provider, conf["transcribe_provider"])
+        self._provider_changed()  # selecting index 0 fires no signal
+        self.cleanup_enabled.setChecked(conf["cleanup_enabled"])
         self.cleanup_model.setCurrentText(conf["cleanup_model"])
         self.cleanup_prompt.setPlainText(conf["cleanup_prompt"] or cfg.default_cleanup_prompt())
         self.transcribe_prompt.setPlainText(conf["transcribe_prompt"])
@@ -443,9 +482,16 @@ class SettingsWindow(QDialog):
         conf["keep_audio"] = self.keep_audio.isChecked()
 
         conf["openai_api_key"] = self.openai_key.text().strip()
-        conf["transcribe_model"] = self.transcribe_model.currentText().strip()
-        conf["cleanup_enabled"] = self.cleanup_enabled.isChecked()
         conf["openrouter_api_key"] = self.openrouter_key.text().strip()
+
+        provider = self.transcribe_provider.currentData() or "openai"
+        self._models[provider] = self.transcribe_model.currentText().strip()
+        conf["transcribe_provider"] = provider
+        for key, name in (("openai", "transcribe_model"),
+                          ("openrouter", "openrouter_transcribe_model")):
+            conf[name] = self._models[key].strip() or cfg.DEFAULTS[name]
+
+        conf["cleanup_enabled"] = self.cleanup_enabled.isChecked()
         conf["cleanup_model"] = self.cleanup_model.currentText().strip()
 
         # Store an empty prompt when it matches the default, so switching the
@@ -477,6 +523,54 @@ class SettingsWindow(QDialog):
 
     # ---- api helpers -----------------------------------------------------
 
+    def _provider_changed(self):
+        """Swap the model box over to the newly chosen provider's own model."""
+        if self._shown_provider:
+            self._models[self._shown_provider] = self.transcribe_model.currentText().strip()
+        provider = self.transcribe_provider.currentData() or "openai"
+        self._shown_provider = provider
+        self.transcribe_model.clear()
+        self.transcribe_model.addItems(TRANSCRIBE_MODELS[provider])
+        self.transcribe_model.setCurrentText(self._models[provider])
+        self.transcribe_status.setText("")
+        self.transcribe_note.setText(
+            t("Runs on OpenRouter, with the key above. Word hints reach the "
+              "cleanup model only.")
+            if provider == "openrouter" else
+            t("Runs on OpenAI, with the key above.")
+        )
+
+    def _load_transcribe_models(self):
+        """The model list of whichever provider is selected."""
+        provider = self.transcribe_provider.currentData() or "openai"
+        self.refresh_transcribe_models.setEnabled(False)
+        self.transcribe_status.setText(t("Fetching model list…"))
+        openai_key = self.openai_key.text().strip() or self.conf.openai_key()
+        openrouter_key = self.openrouter_key.text().strip() or self.conf.openrouter_key()
+        base = self.conf["openai_base_url"]
+
+        def work():
+            try:
+                models = (api.openrouter_models(openrouter_key, transcription=True)
+                          if provider == "openrouter"
+                          else api.openai_models(openai_key, base))
+                self._transcribe_models_loaded.emit(models, "")
+            except api.ApiError as exc:
+                self._transcribe_models_loaded.emit([], str(exc))
+
+        threading.Thread(target=work, daemon=True).start()
+
+    def _on_transcribe_models_loaded(self, models, error):
+        self.refresh_transcribe_models.setEnabled(True)
+        if error:
+            self.transcribe_status.setText(t("Could not fetch the list: {error}", error=error))
+            return
+        current = self.transcribe_model.currentText()
+        self.transcribe_model.clear()
+        self.transcribe_model.addItems(models)
+        self.transcribe_model.setCurrentText(current)
+        self.transcribe_status.setText(t("{count} models loaded.", count=len(models)))
+
     def _load_models(self):
         self.refresh_models.setEnabled(False)
         self.models_label.setText(t("Fetching model list…"))
@@ -520,7 +614,7 @@ class SettingsWindow(QDialog):
 
     def _test_openrouter(self):
         self.or_test_button.setEnabled(False)
-        self.models_label.setText(t("Trying…"))
+        self.or_test_label.setText(t("Trying…"))
         key = self.openrouter_key.text().strip() or self.conf.openrouter_key()
 
         def work():
@@ -533,7 +627,7 @@ class SettingsWindow(QDialog):
 
     def _on_or_test_done(self, ok, message):
         self.or_test_button.setEnabled(True)
-        self.models_label.setText(("✓ " if ok else "✗ ") + message)
+        self.or_test_label.setText(("✓ " if ok else "✗ ") + message)
 
     def _on_test_done(self, ok, message):
         self.test_button.setEnabled(True)

worker.py

@@ -60,13 +60,12 @@ class Pipeline(QObject):
 
         try:
             self.stage.emit(t("Transcribing…"))
+            target = conf.transcribe_target()
             raw = api.transcribe(
+                target,
                 wav_path,
-                conf.openai_key(),
-                model=conf["transcribe_model"],
                 language=conf["language"],
                 prompt=conf["transcribe_prompt"],
-                base_url=conf["openai_base_url"],
             )
 
             if conf["filter_hallucinations"] and vad.looks_like_hallucination(raw, duration):
@@ -107,7 +106,7 @@ class Pipeline(QObject):
                 "ts": time.strftime("%Y-%m-%d %H:%M:%S"),
                 "duration": round(duration, 1),
                 "elapsed": round(time.monotonic() - started, 1),
-                "model": conf["transcribe_model"],
+                "model": target.model,
                 "cleanup_model": conf["cleanup_model"] if conf["cleanup_enabled"] else "",
                 "cleanup_error": warning,
                 "raw": raw,