Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

A podcast refuses on a tight machine rather than voicing at a smaller chunk

オープン
  • 何か問題がありますか？
#1,982 コメント 1 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

@Yogi-cy がすでに取り組んでいます。

2026年10月5日 から。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
45/100
issue の種類
機能追加
明瞭さ
おおむね明確
活発さ
活発
技術スタック
python
領域
backend

調査の方向性

Read the audio.cpp request and session options first to establish how chunk size is configured, then trace check_voicing_memory() in memory.py through _wait_for_memory() and _voice_each() in speech.py. Start with a failing test in test_speech.py; handle models with no chunk_steps and update the catalog documentation. Verify real-model behavior with test_audiocpp_voicing.py and listen to a two-speaker podcast at each enabled step.

索引モデルが issue の本文から書いたものです。

説明

desktop enhancement help wanted

The gap

docs/architecture/local-models/catalog.md lists this under Known gaps:

The audio block's chunk_steps are committed but nothing reads them: short of memory at the default chunk, a podcast refuses rather than stepping down, until a listening test clears the smaller chunks.

Why it matters

Kokoro peaks at 2,347 MB while voicing at the default chunk, and at 1,442 MB at 120 characters and 956 MB at 60. Those numbers were measured on audio.cpp v0.8.2, reviewed and committed. A machine with two and a half gigabytes free is refused a podcast it could produce at half the memory, and the refusal instead suggests a different, lighter model — a worse answer than the same model voicing more slowly.

The manifest rule this page states is that an entry carries only fields some engine reads, precisely so nothing is reviewed and trusted while doing nothing. chunk_steps has been that field since it was added, and the memory check is one argument away from using it.

Where

What done looks like

  • A machine short of memory at the default chunk voices at the largest committed step that fits, rather than refusing, and a model with no steps still refuses with the sentence it gives today.
  • The refusal, when it comes, accounts for the smallest step, not the default peak — otherwise the numbers in the message are wrong.
  • The listening test is part of this issue, not a follow-up. Voice a two-speaker podcast at each step you enable and confirm the audio holds up; say in the PR what you listened to and on which audio.cpp build. A step that degrades the audio should not be enabled however well it fits.
  • Delete the Known gaps line from docs/architecture/local-models/catalog.md in the same PR, and update the chunk_steps bullet under The manifest so it describes a field that is read.

How to verify

surfsense_local/backend/tests/unit/llm/providers/audiocpp/test_speech.py drives voicing against a mock transport with an injectable available callable, so every memory branch is a unit test with no audio involved. tests/packaging/test_audiocpp_voicing.py is the real-model harness, gated on SURFSENSE_TEST_AUDIO_MODELS. The repo is TDD: write the failing unit test first.

cd surfsense_local/backend
uv run pytest -m unit
uv run pytest -m integration
uv run ruff check .

Before you start

Read CONTRIBUTING.md, the local model catalog doc and the Studio doc, which owns the podcast job this serves. Say up front whether you can run audio.cpp with a real model; without that the listening test cannot be done and the issue cannot be finished. Open your PR against dev, not main. Comment here to claim this.

主要言語
Python
スター
16.3k
フォーク
1.6k
平均マージ
12時間 13分
マージ済み PR（30日）
274

環境構築

このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

MODSetter/SurfSense のほかの issue

MODSetter/SurfSense の issue をすべて見る

似ている issue

Python の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。