Tonwerk
wer benchmark
Texts are normalised (NFKC, optional lower-casing and punctuation removal, whitespace collapsed) and aligned with a word-level Levenshtein alignment (unit costs). WER = (S + D + I) / N with N reference words; MER = (S + D + I) / (H + S + D + I); WIP = (H / N) · (H / P) with P hypothesis words; WIL = 1 − WIP (Morris, Maier and Green 2004, as in jiwer). CER is the character Levenshtein distance over the normalised texts including spaces, divided by the reference length. For a corpus the counts.
POST https://ton.halowerk.com/v1/wer-benchmarkInputs
| Field | Type | Required | Description |
|---|---|---|---|
| reference | string | No | Reference (ground-truth) transcript. |
| hypothesis | string | No | Transcript to score. |
| pairs | array | No | Alternative: a corpus of reference/hypothesis pairs (at most 50000 characters in total). |
| lowercase | boolean | No | |
| remove_punctuation | boolean | No | |
| include_alignment | boolean | No | Return the word alignment (single pair only, at most 500 operations). |
Show full input schema
{
"type": "object",
"additionalProperties": false,
"properties": {
"reference": {
"type": "string",
"minLength": 1,
"maxLength": 10000,
"description": "Reference (ground-truth) transcript."
},
"hypothesis": {
"type": "string",
"maxLength": 10000,
"description": "Transcript to score."
},
"pairs": {
"type": "array",
"minItems": 1,
"maxItems": 100,
"items": {
"type": "object",
"additionalProperties": false,
"required": [
"reference",
"hypothesis"
],
"properties": {
"id": {
"type": "string",
"maxLength": 64
},
"reference": {
"type": "string",
"minLength": 1,
"maxLength": 10000
},
"hypothesis": {
"type": "string",
"maxLength": 10000
}
}
},
"description": "Alternative: a corpus of reference/hypothesis pairs (at most 50000 characters in total)."
},
"lowercase": {
"type": "boolean",
"default": true
},
"remove_punctuation": {
"type": "boolean",
"default": true
},
"include_alignment": {
"type": "boolean",
"default": false,
"description": "Return the word alignment (single pair only, at most 500 operations)."
}
}
}Published input example
{
"reference": "the cat sat on the mat",
"hypothesis": "the cat sit on mat"
}Result & documentation
Machine-readable documentation of this hub
Source of this page: the published HALOWERK catalog. Updated: 2026-11-17 UTC.
More endpoints in this hub
- Tonwerk0.01 USDC
accessible transcript html
POSThttps://ton.halowerk.com/v1/accessible-transcript-htmlParses SRT or WebVTT (voice tags <v Name>, "NAME:" prefixes and ">>" speaker changes) or a segments array, merges consecutive cues of the same speaker into turns (a new paragraph after a pause longer than paragraph_gap_s), and writes semant
- Tonwerk0.01 USDC
ad disclosure audio
POSThttps://ton.halowerk.com/v1/ad-disclosure-audioDecodes the complete supplied clip to mono 16 kHz using resource-limited ffmpeg, checks for speech with Silero VAD v6.2 (MIT), and transcribes locally with the pinned Whisper base ONNX model (MIT / Apache-2.0). Conservative English phrase r
- Tonwerk0.005 USDC
ad insertion markers
POSThttps://ton.halowerk.com/v1/ad-insertion-markersRuns ffmpeg 6.1 silencedetect on the episode audio (base64 or URL, at most 10 MiB) and builds mid-roll candidates from internal silences and from Podcasting 2.0 chapter boundaries (a boundary inside a silence or snapped to the nearest silen