Add cloud voice model support for iOS

This commit is contained in:
James Babcock committed 2026-10-10 22:56:07 -07:00
1 parent 5df873e5b1
commit 920cce6a26
24 files changed
+518 -176

No files matched your search

+7 -3
View File
@@ -785,8 +785,10 @@ export const ringConnectionModeSetting = new ConfigSettingEnum<RingConnectionMod
// below now says "OpenAI" to tell the two apart in the picker.
export type VoiceProvider = "onboard" | "onboard-whisper" | "elevenlabs" | "whisper" | "soniox";
// On iOS "onboard" is Apple's on-device recognizer, and there is no
// "onboard-whisper" (the sherpa-onnx models are Android-only).
const voiceProviderLabels: Record<VoiceProvider, string> = {
onboard: "On-device (Moonshine)",
onboard: global.isIOS ? "On-device (Apple)" : "On-device (Moonshine)",
"onboard-whisper": "On-device (Whisper)",
elevenlabs: "ElevenLabs",
whisper: "OpenAI (Whisper)",
@@ -798,7 +800,7 @@ export const voiceProviderSetting = new ConfigSettingEnum<VoiceProvider>({
label: "Transcription Provider",
storageKey: "voice.provider",
defaultValue: "onboard",
values: ["onboard", "onboard-whisper", "elevenlabs", "whisper", "soniox"],
values: global.isIOS ? ["onboard", "elevenlabs", "whisper", "soniox"] : ["onboard", "onboard-whisper", "elevenlabs", "whisper", "soniox"],
formatValue: (value) => voiceProviderLabels[value] ?? value,
isDisabled: (value) => {
if (value === "elevenlabs") return elevenLabsApiKeySetting.get().trim().length === 0;
@@ -806,7 +808,9 @@ export const voiceProviderSetting = new ConfigSettingEnum<VoiceProvider>({
if (value === "soniox") return sonioxApiKeySetting.get().trim().length === 0;
return false;
},
description: "Speech-to-text engine for voice input. ElevenLabs, OpenAI, and Soniox are cloud services that need an API key, with significantly better accuracy than on-device transcription. The two On-device options need their voice model downloaded (below) and never leave the phone.",
description: global.isIOS
? "Speech-to-text engine for voice input. ElevenLabs, OpenAI, and Soniox are cloud services that need an API key (Settings > API Keys), with significantly better accuracy than on-device transcription. On-device uses your iPhone's speech language and never leaves the phone."
: "Speech-to-text engine for voice input. ElevenLabs, OpenAI, and Soniox are cloud services that need an API key, with significantly better accuracy than on-device transcription. The two On-device options need their voice model downloaded (below) and never leave the phone.",
});
const wakeWordActionLabels: Record<WakeWordAction, string> = {
+3 -4
View File
@@ -229,10 +229,9 @@ function settingsSections(): SettingsSection[] {
const deferred = new Set(["Watch"]);
return sections.map(section => {
if (section.label === "Developer") return { ...section, items: [toggleSettingMenuItem(showBleBandwidthSetting)] };
if (section.label === "Voice") return { label: "Voice", items: [enumSettingMenuItem(wakeWordActionSetting), {
label: "On-device dictation (Apple)", disabled: true, onSelect: () => {},
description: "Uses the glasses microphone and your iPhone's speech language. No transcription API key needed. Say Hey Even for hands-free input, or open Voice from the menu and click when finished.",
}] };
// No downloadable on-device models: iOS's on-device option is Apple's recognizer.
if (section.label === "Voice") return { label: "Voice", items: [enumSettingMenuItem(wakeWordActionSetting),
enumSettingMenuItem(voiceProviderSetting)] };
if (deferred.has(section.label)) return { label: section.label, items: [{
label: "Not available on iOS yet", disabled: true, onSelect: () => {},
description: `${section.label} integration has not been ported to iOS.`,