An audio model task map: recognition, synthesis, and generation
Begin an audio model record with its input and expected output. Speech recognition, speech synthesis, audio understanding, and general audio generation answer different questions; one audio label does not make them interchangeable.
Define the task boundary
| Your intended task | Input | Expected output |
|---|---|---|
| Speech recognition | A speech recording | A transcript |
| Text-to-speech | Text and optional documented controls | Generated speech |
| Audio question answering | Audio and a question | An answer about the recording |
| Text-to-audio | A written description | An audio clip matching the request |
These are working definitions for organizing research. A particular model may combine tasks, restrict inputs, or expose different interfaces. Check its own model card before assigning a capability.
Record conditions before comparing
Note the language, recording conditions, clip length, sample rate, channel count, and whether interaction must stream. For speech generation, record whether a reference voice is required, optional, or unsupported. Do not assume that a demo exposes every condition used to produce it.
For synthetic speech, distinguish intelligibility, pronunciation, speaker similarity, prosody, and runtime behavior. A result on one dimension does not establish the others.
This page proposes a vocabulary and contains no model ranking. Use the model-card checklist to verify a concrete candidate.