Text-to-speech generation

Reading a script aloud in a natural voice is the main job. When text is ready, an agent sends it to LMNT, chooses the voice the account allows, and receives spoken audio back. It then applies an approved rule to store or attach that audio where it is needed, a notification, a narration, or a message. Scripts that exceed a length the rule permits, contain characters the voice cannot render, or carry content marked sensitive are stopped before synthesis, and a person decides whether to trim the script, pick another voice, or skip the clip entirely.

Custom voice selection

Matching the right voice to each message keeps output consistent. Reading a label on the task or record, the agent picks from the custom and stock voices the LMNT account offers, then generates the audio for that specific use. The voice chosen is noted on the record so later clips stay uniform. When a task calls for a voice the account has not set up, or the label is ambiguous about which one fits, the agent pauses and asks a person to confirm the choice rather than defaulting to a voice that may sound wrong for the audience.

Audio handoff

Getting finished audio to the right place is where the agent closes the loop. After LMNT returns a clip, the agent reads the destination from an approved rule and files the audio into the matching folder, record, or outgoing message. It records the clip length and where it landed so nothing is generated twice. Clips that come back silent, run shorter than the script suggests, or cannot be written to their destination are left unsent and reported to a person, who checks the source text or the storage setting before the audio goes out.