---
type: "post"
id: "breeze-tts-2-cpp-v0-1-7-adds-acd0c1"
slug: "breeze-tts-2-cpp-v0-1-7-adds-acd0c1"
title: "Breeze-TTS-2.cpp v0.1.7 adds an opt-in voice-call demo"
description: "HoppouAI’s Breeze-TTS-2.cpp v0.1.7 adds an opt-in voice-call demo that connects speech input, streamed text-to-speech, interruption handling and optional screen sharing."
retrieval_nugget: "HoppouAI’s Breeze-TTS-2.cpp v0.1.7 adds an opt-in voice-call demo that connects speech input, streamed text-to-speech, interruption handling and optional screen sharing. For builders, it is a practical example of composing local speech generation with a separate chat server, while making the network and image-input constraints explicit."
published_at: "2026-11-16"
updated_at: "2026-11-16"
record_date: "2026-11-16"
date_kind: "published_at"
topics: ["generative-media","multimodal"]
entities: ["HoppouAI","Breeze-TTS-2.cpp","Tyger","LM Studio","Ollama","llama.cpp"]
related_patterns: ["agent-ready-software-exposes-capabilities"]
novelty_label: "new_fact"
what_changed: "Version 0.1.7 adds an opt-in voice-call demo, letting users speak with a character that responds aloud."
prior_refs: ["https://newruntime.com/posts/hoppouai-released-breeze-tts-2-cpp-gguf-quants-f5d439/"]
source_url: "https://github.com/HoppouAI/Breeze-TTS-2.cpp/releases"
source_urls: ["https://github.com/HoppouAI/Breeze-TTS-2.cpp/releases"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:breeze-tts-2-cpp-v0-1-7-adds-acd0c1"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/breeze-tts-2-cpp-v0-1-7-adds-acd0c1/","markdown":"https://newruntime.com/posts/breeze-tts-2-cpp-v0-1-7-adds-acd0c1.md","json":"https://newruntime.com/posts/breeze-tts-2-cpp-v0-1-7-adds-acd0c1.json"}
markdown_url: "https://newruntime.com/posts/breeze-tts-2-cpp-v0-1-7-adds-acd0c1.md"
---

# Breeze-TTS-2.cpp v0.1.7 adds an opt-in voice-call demo

## Retrieval answer

HoppouAI’s Breeze-TTS-2.cpp v0.1.7 adds an opt-in voice-call demo that connects speech input, streamed text-to-speech, interruption handling and optional screen sharing. For builders, it is a practical example of composing local speech generation with a separate chat server, while making the network and image-input constraints explicit.

## Summary
HoppouAI’s Breeze-TTS-2.cpp v0.1.7 adds an opt-in voice-call demo that connects speech input, streamed text-to-speech, interruption handling and optional screen sharing. For builders, it is a practical example of composing local speech generation with a separate chat server, while making the network and image-input constraints explicit.

## What happened
The GitHub release notes for Breeze-TTS-2.cpp v0.1.7 say the server can host a voice-call demo at `/talk`. A user chooses a character and talks aloud; the character responds in speech. The demo is opt-in: `--talk` enables it, so existing setups do not change. The notes give a launch example that runs `tyger-server` with `models/tyger-q8_0.gguf` on port 8765, then starts `breeze-server` with a Q8_0 Breeze model and `--llm http://127.0.0.1:1234/v1`. The default chat endpoint is that local OpenAI-compatible URL; the default Tyger endpoint is `ws://127.0.0.1:8765/ws`.

The demo uses tyger-server to hear speech and determine when the user has finished speaking. Replies come from an OpenAI-compatible chat server, with LM Studio, Ollama and llama.cpp’s server named as examples. Text streams into speech, and audio begins at the first comma rather than waiting for the full sentence. Users can interrupt the character. The release notes say the system stops playback and puts only the portion actually heard into chat history; if the user only says “mm-hmm” or says nothing, it resumes where it stopped.

The page also supports brief listening sounds such as “mm-hmm” and “yeah.” These can be fixed phrases or generated by the chat model based on what the user is saying. Screen sharing is available through a button in the call. While sharing, a fresh screenshot accompanies everything the user says. That feature requires an image-capable chat model; if the selected model cannot take images, the page informs the user and the character answers without them.

Characters are configured on the page with a name, personality, opening line and voice. A voice can be saved or specified by a description, and the page can clone a new voice from a file or recording. The notes say character configuration is kept in the browser. They identify Windows 10/11 x64 and Linux x64 prebuilt Vulkan downloads, with CPU fallback, and state that AVX2 is required. Using the demo from another device requires both servers to listen on `0.0.0.0`; microphone access then requires HTTPS. The server has no authentication, so the release notes advise keeping it on `127.0.0.1` unless another layer handles protection.

## What the archive adds
New Runtime covered [HoppouAI’s release of Breeze-TTS-2.cpp](https://newruntime.com/posts/hoppouai-released-breeze-tts-2-cpp-gguf-quants-f5d439/) on 2026-11-03, noting its GGUF quantizations and dedicated C++/ggml runtime for BreezeBlue’s 3B TTS model. This is a new capability on that known project, not a restatement of the initial runtime release.

## Trend signal
This release offers a small example of agent-ready software exposing capabilities through a web interface and server endpoints, but it is a voice demo rather than a clear case of external agents composing permissioned capabilities. The connection is therefore tentative; it does not yet establish the broader pattern described in [Agent-ready software exposes capabilities](https://newruntime.com/patterns/agent-ready-software-exposes-capabilities/).

## Why now
The release does not explain why the voice-call demo arrived at this moment, so there is no supported timing claim to make.

## Why it matters
For teams prototyping voice interfaces, the demo assembles several useful behaviors without requiring a single all-in-one model: one service handles speech timing, a compatible chat server produces replies, and Breeze-TTS-2.cpp streams the voice. Interruptions are treated as conversation state, not just cancelled audio: only what the caller heard is retained. That distinction matters when a voice assistant must recover naturally after being cut off.

The screen-sharing option also makes the interaction multimodal, but its requirement for an image-capable chat model creates a clear dependency to test. The no-authentication warning is equally consequential. A local demo can be useful for experimentation, but exposing it to a network without a protective layer would create avoidable risk. Builders can use the release as a prototype reference while keeping the network boundary and model capabilities explicit.

## Limits and uncertainty
The release notes describe the demo’s behavior but do not report independent evaluations of speech recognition quality, interruption latency, response reliability or the accuracy of screenshot interpretation. Performance will depend on the selected chat server, image-capable model and local hardware. The release notes also do not establish whether the demo is intended for production deployment; the explicit lack of authentication means it should not be treated as a network-ready service without additional protection.

## Skill to build now
Try the documented local setup with separate Tyger and OpenAI-compatible chat servers, keeping both bound to localhost. Test three cases: interrupt the character mid-sentence, speak only a short acknowledgment, and share a screen with an image-capable model. Check whether the retained conversation matches what was actually audible, and verify that the no-image fallback is clear before considering a broader deployment.

## Reader takeaway
Breeze-TTS-2.cpp v0.1.7 turns its local speech runtime into an opt-in voice-call prototype, including interruption-aware history and optional screen context. The useful lesson is in the composition and its boundaries: separate speech and chat components can work together, but image support, HTTPS for remote microphone use and protection for an unauthenticated server must be planned deliberately.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Generative Media](https://newruntime.com/topics/generative-media.md): Explore the generative-media topic hub.
- topic hub: [Multimodal](https://newruntime.com/topics/multimodal.md): Explore the multimodal topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)
