Alma voice covers two flows: speech-to-text (record a voice message and Alma transcribes it before sending to the chat) and text-to-speech (the AI reads its responses aloud). Both work on every plan that has chat access and run inside the unified monthly budget — no separate voice subscription, no per-minute counter visible to users.
Speech-to-text powered by Deepgram Nova-2 and text-to-speech by ElevenLabs. Both are optional services with graceful degradation. All voice interactions maintain full persistent memory context.