readio

readio is a terminal text to speech tool. It plays local speech with PyKokoro, renders bounded memory WAV files, and publishes completed audio through the external save-to-spotify CLI.

Install

python -m pip install -e ".[cpu]"

For GPU ONNX Runtime:

python -m pip install -e ".[gpu]"

PyKokoro may download model and voice assets on first use. Spotify publishing requires the separately installed save-to-spotify executable and its authenticated session. Readio never reads Spotify credential files.

Playback

readio speak "Hello from the terminal."
printf '%s\n' "Read this from stdin." | readio speak
readio speak --file notes.md
readio speak --file notes.md --select last-paragraph
readio speak --file notes.md --select paragraph:3
producer-command | readio speak --live

Configuration

Initialize one user-owned Readio configuration and its storage:

readio config init
readio config show
readio config validate
readio config set reader.voice bf_emma
readio config set voices.kokoro.roles.analyst am_michael
readio config set ssmd.voice_provider kokoro

The default configuration uses platformdirs for the config, template, ingest, and output locations. READIO_CONFIG overrides the config file path. Existing legacy files containing only [reader] continue to load and are upgraded to schema 1 when saved.

The configuration contains reader settings, SSMD defaults, provider-specific voice IDs, and logical role bindings. Templates refer to roles such as host, analyst, guest, and narrator, while ordinary literal text continues to use reader.voice.

Templates

Built-in templates are copied into the user template directory during initialization. They are user-owned and are not overwritten by normal initialization or package upgrades.

readio template path
readio template list
readio template show podcast
readio template add custom --file custom.ssmd
readio template remove custom
readio template reset podcast
readio template reset --all

Create an agent-editable draft with an automatic filename:

draft="$(readio template use podcast)"

The returned path is under the configured ingest directory. A caller can request a filename with readio template use podcast --name weekly-review.ssmd.

Ingest directory

The ingest directory stores text, Markdown, and SSMD files created for later processing.

readio ingest path
readio ingest new
readio ingest new --name notes.txt
readio ingest new --template podcast --name episode-42.ssmd
readio ingest list

Automatic names contain a UTC artifact ID such as 20260824T111423Z-5f8ab31c. Explicit names are relative to the ingest directory and path traversal is rejected.

Automatic WAV output

The output path is optional:

readio render "Hello from a file."
readio render --file "$draft"
readio render --file "$draft" -o episode.wav

Without -o, Readio uses the configured output directory. Generated ingest files retain their artifact stem in the WAV name. Other input files receive a new artifact suffix, and literal text uses the readio prefix. Automatic names never overwrite an existing file. Explicit output remains atomic and requires --force for replacement.

SSMD consumption and authoring checks

For .ssmd inputs, Readio parses the document through SSMD 0.8.3 and passes a PyKokoro SSMDRenderConfig containing only missing Readio role defaults. Document voice_bindings remain authoritative. Normal speak, render, and spotify commands do not invoke ssmd create, rewrite the source, or require generic round-trip validation.

Inspect a document before rendering:

readio ssmd check episode.ssmd
readio ssmd check episode.ssmd --json
readio ssmd check episode.ssmd --roundtrip

readio template validate --all checks shipped or configured templates with the same consumer preflight. Add --roundtrip for strict SSMD authoring validation. Unknown logical roles fail before model inference with a Readio diagnostic.

Spotify publishing

readio spotify --file "$draft" --title "Weekly Review" --wait

Publishing is explicit. Without --output, Readio renders to a secure temporary WAV and deletes it after upload or failure. With --output, it retains and uploads that file. Readio invokes save-to-spotify --json and does not inspect credentials or perform authentication.

Doctor

readio doctor

Doctor is offline. It reports Readio configuration, configured directories and their existence, PyKokoro, SSMD module and executable availability, the selected provider, voice IDs, logical roles, sound dependencies, and save-to-spotify. It does not create directories, modify configuration, inspect credentials, or call the network.

Agent Skill

The portable skill is in skill/readio/SKILL.md. It uses Readio templates and commands directly. It does not teach raw SSMD voice discovery, create, lint, temporary file management, or manual cleanup for normal podcast workflows.