Dictaria question classifier
Detects whether the latest chunk of live meeting speech asks something the listener should answer. Dictaria runs it on the user's device (browser or desktop app) with onnxruntime-web: no text leaves the device to be classified.
Labels
question, request, social, rhetorical, statement, fragment (in labels.json order).
Dictaria treats question + request probability >= 0.5 as "answer this".
question: a real question someone else is expected to answer.request: asks for information without question form ("walk me through...", "cuéntame...").social: comprehension checks, tag questions, small talk, logistics, action requests.rhetorical: the speaker answers it or says it for emphasis.statement: asks nothing.fragment: cut off mid-thought or filler only.
Model
- Base:
paraphrase-multilingual-MiniLM-L12-v2(Apache 2.0) with a 6-way classification head. - Fine-tuned on about 3,820 synthetic meeting-speech examples (90% training, 10% validation) in English, Spanish, French, German and Portuguese, labeled by an AI model.
- Exported to ONNX and quantized to int8 per channel (
model.onnx, 118 MB). - Input: at most 96 tokens (BOS + the last 94 tokens of the speech + EOS).
Results
340 independently written test cases, never used for training or tuning, scored through the browser code path with the int8 model:
| Language | Accuracy |
|---|---|
| English | 97.5% (78/80) |
| Spanish | 96.3% (77/80) |
| French | 95.0% (57/60) |
| German | 96.7% (58/60) |
| Portuguese | 98.3% (59/60) |
| All | 96.8% (329/340) |
Speed: 12 to 30 ms per check in a browser worker (wasm, one thread).
Limitations
- Judges only the text it is given; it does not know whether a question was already answered.
- Trained on synthetic speech, not real transcripts; speech-recognition errors in real meetings may lower accuracy.
- Only the five languages above were tested.