On-device voice on iPhone without the cloud: open-source app combines Phonon TTS and Pipecat
An open-source iOS app runs speech-to-text and text-to-speech entirely on the phone, using embedded Pipecat, Apple's SpeechAnalyzer and the Gradium Phonon 31M model, with no network inference.
Source: Little Voices: app de voz on-device com Phonon TTS e Pipecat no iPhone (github.com). Text prepared with AI from this source.
What happened and what to do
The open-source Little Voices app, published in October 2026, is a voice game in which an angel and a devil on your shoulder answer spoken questions. It uses embedded Pipecat in Python to orchestrate the pipeline, Apple's SpeechAnalyzer to convert speech to text, and the Gradium Phonon 31M model to synthesize the voice. All processing happens on the iPhone itself, with no network inference calls, and the pipeline code can be inspected publicly.
For companies, the case shows that part of the voice functions can run on the device, which reduces exposure of customer and employee audio, lowers dependence on connectivity, and can cut latency and recurring API costs. In practice, the first step is to map which voice flows touch sensitive data and which need large models that do not fit on the device. Then it is possible to prototype a local pipeline with speech-to-text and TTS in a restricted use case, measure latency and battery use on real devices, and define when inference stays local and when it returns to the company's private infrastructure.
How the consultancy can help
Wendelmaques can diagnose where on-device voice processing makes sense for your product or process, assessing privacy, latency, battery life and device compatibility, and then implement a local voice pipeline with monitoring, within a defined scope.
Next step
If your company is evaluating voice on phones or in field applications, send a short description of your case and receive a scoped proposal.
Let's assess your project
I reply with scope and a proposal.