The State of Voice Interactions on Android
summary
Android voice stacks break into four concrete pieces: wake word detection, speech-to-text, NLP/NLU and text-to-speech, and Android exposes SpeechRecognizer and TextToSpeech APIs since 2009–2010 plus Dialogflow for intent parsing. App Actions let Google Assistant open app screens with parameters but are underused; Gemini API can replace STT+NLP, Gemini Nano and Gemma (on-device LLM) enable offline inference, and third-party SDKs (Picovoice demoed full offline timers and on-device LLM replies with zero network traffic) enable cross-platform, privacy-first builds.
AI-written summary · about
chapters (20)
- 0:00Voice interactions on Android
- 1:05Why voice matters today
- 4:15Four building blocks of voice
- 8:36LLMs vs traditional NLP
- 11:30Three strategies to integrate voice
- 13:35First-party Android voice APIs
- 16:47Voice Access accessibility
- 17:50App Actions with Assistant
- 19:37Assistant to Gemini transition
- 21:25Gemini API and Gemma offline
- 25:44Third-party voice SDKs
- 27:34Cross-platform voice solutions
- 29:43In-house open-source AI
- 31:54Live demos of voice apps
- 35:49On-device LLM demo with Pico Voice
- 36:10Key takeaways for voice apps
- 37:36Creative voice use cases
- 39:21Small models for niche voice
- 40:45Future of voice integration
- 41:06Closing remarks and Q&A
related talks
43mThe State of Voice Interactions on Android
47mBuilding Voice-First Android Apps- 47mBuilding Voice-First Android Apps
29mAndroid Gets a Voice: Building Real-Time Agents
35mHaving a Voice on Android: How hard can it be?
20mHow Google Assistant's architecture powers voice features in your apps
55mConversational power for Android apps
16mHow to voicify your Android app