SolarOS 4.17.7 manual · job

speechd

Offline text-to-speech queue. The optional job loads a PicoTTS voice from storage into PSRAM and accepts asynchronous requests from native applications, Python, and Lua.

job start speechd /voices/en-US
say "Solar O S is ready"
say --pitch 85 --speed 120 "Solar O S is ready"
say --file /documents/announcement.txt
say --file /books/novel.txt
job status speechd
job stop speechd

say --file <path> validates and reads a plain UTF-8 text file in bounded chunks. It keeps the shell in a foreground playback mode, but advances from shell events so the display and progress bar refresh while speech is running. The percentage is submitted-file progress: it advances whenever the next bounded chunk is accepted by speechd. The complete file is one streaming speech request: PicoTTS and the audio player stay open between chunks, while a single bounded handoff slot prevents the file from being buffered in memory. Press Esc or Ctrl+C to cancel the current speech request and stop reading. The foreground command returns only after speechd confirms cancellation, so a new say request cannot race the previous stream's cleanup. Cancellation discards queued text and soft-resets PicoTTS instead of synthesizing the unspoken remainder of the file. There is no file-size limit because only one bounded chunk is held and submitted at a time. Tab completion after --file lists filesystem paths. Ordinary say <text...> remains asynchronous and returns the queued request ID. --pitch 50..200 and --speed 20..500 use PicoTTS's native controls; both default to 100 and apply to ordinary text and file playback.

The voice directory must contain ta.bin and sg.bin. No voice blobs are compiled into firmware.bin. The repository's top-level picotts_voices/ directory contains ready-to-copy voices for en-GB, en-US, de-DE, es-ES, fr-FR, and it-IT. Builds also mirror them below .pio/build/<environment>/picotts_voices/. Copy one or more complete locale directories to SD or flash storage, then select the language when starting the job:

job start speechd /voices/de-DE

Restarting speechd with another directory changes the language at runtime. The directory name is used as the voice label in job status speechd; PicoTTS validates the TA and SG resource contents while starting. Missing, truncated, or incompatible resources make the job fail instead of accepting requests.

The job acquires the selected audio output only while an utterance is active and releases it between requests. Normal requests wait for another audio owner; callers can mark disposable notifications drop_if_busy.

The queue holds eight requests of at most 512 UTF-8 bytes. Request IDs support polling and cancellation through solaros.speech. Stopping the job cancels pending speech and releases the roughly 1.1 MiB PicoTTS engine allocation plus the selected TA and SG buffers.