# TokataHQ evidence audit — 24 September 2026

## Scope and limits

Reviewed all 49 existing Tracker records and all 62 Resources listings against first-party documentation, product interfaces, original model/dataset cards, repositories, or papers. Added two separately scoped Gemini API records. There are now **51 product surfaces and 106 capability records**: 17 officially supported, 2 not supported, and 87 unknown. No capabilities were promoted to TokataHQ tested. No new account-based product tests, dataset downloads, or model execution were performed.

Source retrieval failures are not proof that a product or resource is discontinued. Source-reviewed means the listing was inspected, not that software works or its quality is established. Unknown fields are left unverified rather than guessed. [Complete evidence records](evidence-audit.json) contain every per-capability source, last-verified/last-reviewed date, and every resource decision.

## 1. Tracker changes

Every existing record now uses capability-level evidence and dates. Removed product-wide support labels and inference-based confidence grades. Separated transcription, audio translation, text translation, subtitle translation, live captions, and live translation; an Igbo course is no longer treated as an interface language. OpenAI Whisper is explicitly the open-source model, not the hosted API.

- Added **Gemini 3.8 Flash TTS**, model gemini-3.8-flash-tts: officially documented Igbo speech generation in the developer API. No consumer-app, ASR, or translation inference.
- Added **Gemini Live Translation**, model gemini-3.5-live-translate-preview: Igbo absent from the reviewed language table; Unknown, not unsupported.
- **Gemini Flash audio observations** remain Unknown. Three reported successful samples are attributed; no percentage is published without a designed evaluation.
- **Notta**: documented real-time translated text into Igbo is target-only; Igbo source transcription remains Unknown.
- **Windows**: documented Igbo language interface pack is partial localization with an English base; keyboard and speech are separate unknowns.
- **Apple keyboard**: current iOS/iPadOS feature-availability source confirms keyboard input only.
- **VEED, DeepL, Google Translate, Microsoft Translator, Gboard, Google Voice Search, Cloud Speech-to-Text V2/Chirp 2, and ElevenLabs Scribe** retain only feature-specific documented support.
- **YouTube Live** remains Not supported because documentation explicitly limits automatic live captions to English. **Otter transcription** remains Not supported because its expressly exhaustive language statement excludes Igbo. Omission-only findings elsewhere are Unknown.

## 2. Resources added

None. Audited the existing 62 entries instead of expanding the directory.

## 3. Resources removed / flagged

No records removed. Eight are visibly flagged:

- **IGT-004 — Nkọwa okwu Igbo Translate Chrome Extension**: The project link could not be retrieved. This does not prove that the extension has been discontinued.
- **IGT-010 — IgboAPI Dataset Expansion**: Igbo inclusion is not inferred from the repository name. Documentation or a permitted data inspection is needed.
- **IGT-011 — IgboSynCorp**: The Harvard Dataverse landing page could not be read during this review. No duration or downloadable-release claim is retained.
- **IGT-024 — Olu Igbo**: A hackathon project listing is not independent confirmation of a working released app.
- **IGT-034 — JW300 - Igbo**: Retained as a flagged historical lead, not a currently available dataset. The previous download URL is retained as an evidence lead only.
- **IGT-046 — Edémédé**: Code exists; live site retrieval failed. License wording needs clarification.
- **IGT-047 — Igbo Keyboard - Tone Marks & Diacritics**: Store retrieval failure is not proof of discontinuation.
- **IGT-051 — Igbo ASR: MMS Adapter + KenLM**: Model card exists, but no available weight download was verified. Comparative performance claims are not reproduced.

Removed unverified quantities, performance comparisons, unsupported maintenance labels, and unverified origin classifications. Replaced the origin filter with source-review status. Removed redundant or unverified secondary links; retained primary evidence links and explicit flagged historical leads. Corrected FLORES-200 links that previously pointed to Belebele. Clarified Naija-Lex as stopwords; N-ATLaS commercial/user restrictions; NaijaVoices registration and non-commercial license; Soro-TTS and NLLB non-commercial terms; SeamlessM4T Igbo input/text-output scope; Mozilla Igbo TTS access requests; and preliminary IlùBench adjudication.

## 4. Downgraded claims

Former omission-based “Not officially supported” claims and untested “Experimental” behavior are Unknown unless an explicit exclusion is documented. Products affected include: YouTube, Instagram, WhatsApp, Premiere Pro, Firefly Generate Speech, Descript, Canva, Otter, Riverside, Notta, Google Meet, Microsoft Teams, Zoom, Android Live Caption, Apple Live Captions, Siri, Google Assistant, Alexa, Google Translate, Microsoft Translator, Amazon Translate, Apple keyboard, Windows, Google Search, Cloud Text-to-Speech, Amazon Transcribe, Amazon Polly, Azure AI Speech, ElevenLabs voices, OpenAI Whisper (open-source model), Text to Speech API, ChatGPT, Claude, Copilot, Meta AI, Duolingo. This list includes mixed products where only some capabilities changed; the table below records exact before/after capability sets.

## 5. New or corrected primary sources

- [Gemini 3.8 Flash TTS model](https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash-tts)
- [Gemini Live Translation](https://ai.google.dev/gemini-api/docs/live-api/live-translate)
- [DeepL Translator language coverage](https://support.deepl.com/hc/en-us/articles/360019925219-DeepL-Translator-languages)
- [Apple feature availability](https://www.apple.com/ios/feature-availability/)
- [Microsoft Translator language table](https://www.microsoft.com/en-us/translator/languages/)
- [FLORES-200 documentation](https://github.com/facebookresearch/flores/blob/main/flores200/README.md)
- [Unicode Igbo locale source](https://github.com/unicode-org/cldr/blob/main/common/main/ig.xml)
- [IrokoBench paper](https://aclanthology.org/2025.naacl-long.139.pdf)
- [AfriHate paper](https://aclanthology.org/2025.naacl-long.92.pdf)

## 6. Manual TokataHQ testing still needed

All Unknown product features need feature-specific documentation or repeatable testing before a positive/negative conclusion. Priorities: complete the Gemini Flash three-clip record; test captions in creator tools; test live features separately; evaluate ASR, translation direction, and TTS with Igbo speakers; check device/account availability of announced voice input; test search queries and sung pronunciation. Officially supported entries still need quality and dialect evaluation. An available resource listing is not an installed-app, downloadable-weights, or license-compliance test.

## Content and layout

Restored two concrete examples on the homepage and preserved the “A thousand people…” line. Clarified TokataHQ → #PutIgboInTech and experiences → evidence → advocacy. Added an explicitly unpaid volunteer route, retaining a separate product-partnership contact route. Updated Temple’s concise bio and added his collaborator announcement without inventing a joining date. Existing brand colors, typography, portrait dimensions, and navigation structure are preserved.

## Product-by-product change ledger

| Product | Previous capability statuses | Audited capability statuses |
|---|---|---|
| YouTube | captions: no; translation: unclear | captions: unclear; subtitletranslation: unclear |
| Instagram | captions: unclear; translation: unclear; reeltranslation: no; tts: unclear | captions: unclear; translation: unclear; audiotranslation: unclear; tts: unclear |
| Facebook | captions: unclear; tts: unclear; translation: unclear | captions: unclear; tts: unclear; translation: unclear |
| WhatsApp | asr: no; tts: no; translation: no | asr: unclear; tts: unclear; translation: unclear |
| TikTok | captions: unclear; translation: unclear; tts: unclear | captions: unclear; translation: unclear; tts: unclear |
| CapCut | captions: unclear; translation: unclear; tts: unclear | captions: unclear; subtitletranslation: unclear; tts: unclear |
| Premiere Pro | asr: no; captions: no | transcription: unclear; captions: unclear |
| Firefly Generate Speech | tts: no | tts: unclear |
| Descript | asr: no; captions: no; tts: no | transcription: unclear; captions: unclear; tts: unclear |
| VEED | asr: yes; captions: yes; translation: yes | asr: yes; captions: yes; subtitletranslation: yes |
| Canva | captions: no | captions: unclear |
| Otter | asr: no; captions: no | transcription: no; captions: unclear |
| Riverside | asr: no; captions: no | transcription: unclear; captions: unclear |
| Notta | asr: no; translation: partial | transcription: unclear; livetranslation: yes |
| Google Meet | captions: no; translation: no | livecaptions: unclear; livetranslation: unclear |
| Microsoft Teams | captions: no; asr: no; translation: no | livecaptions: unclear; transcription: unclear; livetranslation: unclear |
| Zoom | captions: no; translation: no | livecaptions: unclear; livetranslation: unclear |
| Android Live Caption | captions: no | livecaptions: unclear |
| Apple Live Captions | captions: no | livecaptions: unclear |
| Siri | assistant: no; asr: no; tts: no | assistant: unclear; asr: unclear; tts: unclear |
| Google Assistant | assistant: no; asr: no; tts: no | assistant: unclear; asr: unclear; tts: unclear |
| Alexa | assistant: no; asr: no; tts: no | assistant: unclear; asr: unclear; tts: unclear |
| Google Translate | translation: yes; asr: yes; tts: unclear | translation: yes; asr: yes; tts: unclear |
| DeepL | translation: yes | translation: yes |
| Microsoft Translator | translation: yes; asr: no; tts: no | translation: yes; asr: unclear; tts: unclear |
| Amazon Translate | translation: no | translation: unclear |
| Gboard | input: yes; asr: yes | input: yes; asr: yes |
| Apple keyboard | input: yes; asr: no | input: yes; asr: unclear |
| Windows | input: yes; ui: partial; asr: no | input: unclear; ui: yes; asr: unclear |
| Google Search | ui: yes; search: unclear; asr: yes | ui: unclear; search: unclear; asr: yes |
| Spotify | search: unclear; ui: unclear | search: unclear; ui: unclear |
| Cloud Speech-to-Text | asr: yes | asr: yes |
| Cloud Text-to-Speech | tts: no | tts: unclear |
| Amazon Transcribe | asr: no; captions: no | asr: unclear; captions: unclear |
| Amazon Polly | tts: no | tts: unclear |
| Azure AI Speech | asr: no; tts: no | asr: unclear; tts: unclear |
| ElevenLabs Scribe | asr: yes; captions: yes | transcription: yes; captions: yes |
| ElevenLabs voices | tts: no | tts: unclear |
| OpenAI Whisper (open-source model) | asr: no; captions: no | transcription: unclear; captions: unclear |
| Text to Speech API | tts: no | tts: unclear |
| ChatGPT | aitext: exp; translation: exp | aitext: unclear; translation: unclear |
| ChatGPT Voice | voice: unclear; asr: unclear; tts: unclear | voice: unclear; transcription: unclear; tts: unclear |
| Gemini Flash (reported audio tests) | asr: unclear; translation: unclear; aitext: unclear; voice: unclear | audiotranslation: unclear; transcription: unclear; aitext: unclear; voice: unclear |
| Claude | aitext: exp; translation: exp | aitext: unclear; translation: unclear |
| Copilot | aitext: exp; translation: exp | aitext: unclear; translation: unclear |
| Meta AI | aitext: exp; translation: exp; voice: exp | aitext: unclear; translation: unclear; voice: unclear |
| Suno | singing: unclear | singing: unclear |
| Duolingo | ui: no | course: unclear |
| YouTube Live | livecaptions: no | livecaptions: no |
| Gemini 3.8 Flash TTS | New entry | tts: yes |
| Gemini Live Translation (preview API) | New entry | livetranslation: unclear |

Legacy codes: yes = supported; no = previously “not officially supported”; exp = experimental; partial = limited; unclear = unverified. Audited codes: yes = Officially supported; no = Not supported by explicit exclusion; unclear = Not documented / Unknown.