TalkToMePy TTS (Deprecated)

Deprecation Status

• Status: Deprecated
• Scope: Legacy-only, backward compatibility
• Successor: Use the speech workflow successor in gaelic-ghost/a11y-skills for new automation and active maintenance.
• Guidance: Do not choose this skill for new setups unless you explicitly need this older TalkToMePy-specific behavior.

Use this skill when the user asks to hear text spoken aloud from the local machine.

Post-Invocation Resolution Rules

Apply these rules before synthesis to make speech-source selection deterministic in ambiguous contexts.

Invocation detection

• Treat the skill as invoked when the user clearly calls it, including skill chip/link invocation, $talktomepy-tts, or equivalent direct imperative phrasing.
• Ignore incidental mention in unrelated prose.

Source text precedence

1. Suffix invocation pattern:
   • If invocation appears at the end of the user message, speak the user text before the invocation token.
2. Standalone invocation pattern:
   • If the user message is only the invocation, speak the immediately previous assistant message.
3. Prefix invocation pattern:
   • If invocation appears at the beginning of a longer user message, speak the immediately previous assistant message.
   • After speaking, continue responding to the remaining user message normally.

No-prior-assistant fallback

• Standalone invocation with no previous assistant message:
  • Explain there is no prior assistant message to read.
  • Ask whether the user wants to provide text, or wants current text spoken.
• Prefix invocation with no previous assistant message:
  • Explain the chat has no earlier assistant message.
  • Ask whether to speak the current user text.
  • If the user is upset or confused, explain invocation-placement rules and how to trigger the behavior they want.
• Suffix invocation:
  • Speak the preceding user text even when no prior assistant message exists.

Long-content handling

• Estimate length using approximate whitespace-based word count.
• If selected text is longer than about 250 words, ask before synthesis with choices:
  • Speak full
  • Summarize then speak (recommended)
  • Cancel
• If the user chooses summary, generate a concise summary first, then synthesize the summary.
• If the user chooses cancel, do not synthesize.

User dissatisfaction fallback

• If the user is displeased with the skill's default behavior or capabilities, offer to help adapt their own version.
• Offer options:
  • Fork and customize from gaelic-ghost/productivity-skills.
  • Use $skill-creator to create a custom variant modeled after this skill.
  • Open an issue at repository issues.
  • Prepare and submit a PR at repository pull requests.
  • Contact Gale via GitHub profile to discuss improvements.

Execution order

1. Resolve source text using the rules above.
2. Apply long-content confirmation behavior if needed.
3. Run the existing synthesis flow.
4. Preserve existing load/retry/playback behavior.

What this skill does

• Calls the local TalkToMePy v0.5+ service (/health, /model/load, /model/status, /synthesize/voice-design)
• Handles async model loading behavior (/model/load may return 202)
• Retries synthesis on 503 using Retry-After
• Saves generated WAV output to ./tts_outputs in the current working directory by default
• Plays audio via afplay on macOS

Preconditions

• TalkToMePy service is running (default http://127.0.0.1:8000)
• macOS afplay is available

Default workflow

1. Resolve which text to speak using post-invocation resolution rules.
2. Ensure service is healthy:
   • curl -fsS http://127.0.0.1:8000/health
3. Trigger model load (idempotent):
   • curl -sS -X POST http://127.0.0.1:8000/model/load -H "Content-Type: application/json" -d '{"mode":"voice_design","strict_load":false}'
4. Wait for ready state via /model/status
5. Synthesize + save + play using bundled script:
   • scripts/speak_with_talktomepy.sh --text "..."

Script usage

scripts/speak_with_talktomepy.sh --text "Read this text aloud"

Defaults:

• language: English
• default style: energetic (warm/friendly/brisk feminine-or-androgynous)
• output path: ./tts_outputs/tts-YYYYMMDD-HHMMSS.wav

Style preset flags:

• --style-energetic
• --style-soft
• --style-neutral

Alternative style syntax:

• --style energetic|soft|neutral

Optional flags:

• --instruct "..." fully custom voice/style instruction
• --language English
• --base-url http://127.0.0.1:8000
• --save /path/output.wav custom save path
• --no-play generate only, do not play

Optional env var overrides:

• TALKTOMEPY_BASE_URL
• TALKTOMEPY_OUTPUT_DIR
• TALKTOMEPY_MAX_WAIT_SECONDS
• TALKTOMEPY_MAX_SYNTH_RETRIES
• TALKTOMEPY_DEFAULT_RETRY_AFTER_SECONDS

Automation Templates

Use $talktomepy-tts inside automation prompts so Codex loads the service checks and synthesis guardrails in this skill.

For ready-to-fill Codex App and Codex CLI (codex exec) templates, including unattended-safe defaults (--no-play) and placeholders, use:

• references/automation-prompts.md

References

• Automation prompt templates: references/automation-prompts.md

If synthesis fails, surface HTTP status/body and suggest checking:

• /model/status
• launchd logs: ~/Library/Logs/talktomepy.stderr.log