Audio, speech and transcription
Text-to-speech, speech-to-text and audio in posts with Aimogen Pro, across OpenAI, ElevenLabs, Google and Azure.
Aimogen Pro works with audio in two directions: turning text into speech, and turning speech into text.
Provider credentials are in Speech and voice providers.
Text-to-speech#
In the chatbot#
The chatbot can speak its replies, with a talking avatar option. See Voice, speech and talking avatars.
In posts#
The AI Content Editor can generate an audio or video version of a post and attach it to the content — an audio version of an article, generated automatically.
| Setting | What it does |
|---|---|
Text To Audio/Video Template | What text is spoken. Accepts placeholders |
AI Generated Audio/Video Location | Where the player is inserted |
HTML Text To Prepend Audio/Video | Markup before the player, for a heading or a note |
Copy Audio/Video Files | Store locally rather than hotlinking |
Voice settings mirror the chatbot ones for whichever provider you choose.
Speech-to-text#
Transcription uses OpenAI models — whisper-1, gpt-4o-mini-transcribe or gpt-4o-transcribe. This is
the only transcription backend in the plugin.
Where you can transcribe#
| Context | How |
|---|---|
| Front end | [aimogen-audio-converter] |
| Workflows | The stt_openai OmniBlock |
| Editor templates | The audio-to-text settings |
| Playground | The Speech To Text tab |
In the AI Content Editor#
An editor template can transcribe audio attached to a post and add the text to the content:
| Setting | What it does |
|---|---|
Maximum Number Of Audio Files To Process | Batch limit per post |
AI Model To Use For Speech-to-Text | Which transcription model |
Prompt To Send To The AI | Guides transcription — useful for domain terminology |
AI Temperature | Transcription temperature |
Textual Template To Be Added To The Post | How the transcript is formatted |
AI Generated Text Location | Where it is inserted |
%%audio_to_text%% carries the transcript.
The prompt field is more useful than it looks: supplying expected names, jargon and product terms noticeably improves accuracy on domain-specific audio.
Podcast and video workflows#
The combination that works well:
- Transcribe the recording with speech-to-text.
- Restructure the transcript into an article with an AI text step — spoken language does not read well as prose.
- Publish with the audio embedded.
That is straightforward to build as an OmniBlock workflow: stt_openai, then
ai_text, then save_post.
For YouTube specifically, the YouTube bulk creator already does this using captions, which is cheaper than transcribing the audio.
Usage limits#
Speech has its own credit pool: System & Logs › Text-to-Speech Limits, with the same structure as the text limits — per-user and per-guest credits, a time frame, and absolute or rolling windows.
Configure it before exposing speech to visitors. See Usage limits and credits.
Accessibility#
Audio versions of articles genuinely help some readers. Two things to get right:
- Never autoplay. Provide a player the reader chooses to start.
- Keep the text. Audio should supplement the article, never replace it — text is searchable, translatable, and accessible to deaf readers.
Common problems#
No voices in the dropdown. The provider key is missing or invalid.
Audio is truncated. The text exceeded the provider per-request character limit. Split it.
Transcription is inaccurate. Poor source audio, or unfamiliar terminology. Supply expected terms in the transcription prompt.
Transcription fails on large files. Provider file size limits. Split the audio.
Costs are higher than expected. Speech is per character; transcription is per minute. Set the text-to-speech limits.
Related#
Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.