Ka taea te whakawātea i te āhua o te AI
Ko te AI e whakamahia ana e tātau, i reira e haere ana, me te pēhea e whakaae ana tātau ki ngā whakaritenga whakapuaki (EU AI Act Article 50, e pā ana ki te 2026-08-02).
TL;DR
- Ko ngā tāruatanga katoa e hangaia ana e AI. Ka mau te huaputa i tētahi whakapuakitanga me te pūrere.
- Ko ngā AI katoa e haere ana i runga i tātau ake GPU. Kāore e tuku ana mātou i tō tātau oro, tuhituhi rānei ki OpenAI, Anthropic, Google, i tētahi atu LLM API tuatoru rānei.
- Ehara i te whakaako i tētahi tauira i runga i tōtou whakahua. He tauira pūtake tūwhera ia tauira e whakahaeretia ana e tātau. Ko ngā whakarerekētanga e mahia ana e koe e kohia ana anake mēnā e kōwhiria ana e koe.
- Ko ngā tārua reo whakairo (TTS) e tohu ana he AI-mārō i roto i te ingoa faila, i te metadata, i te pātū.
Kāhua e whakamahia ana e tātau
| Kāhua | Ka whakamahia mo | Whakawhiwhinga | Ka haere i runga |
|---|---|---|---|
| Whisper large-v3-turbo (faster-whisper) | Whakawaehere (whakahaere) | MIT | Ko tātau GPU |
| Whisper large-v3 (faster-whisper) — branded “STT.ai Enhanced” | Whakawaehere (whakahaere) | MIT | Ko tātau GPU |
| Vosk | Ka puta te kupu i te wā tūturu | Apache 2.0 | Ko tātau GPU |
| SpeechBrain ECAPA-TDNN | Whakahauhautanga kaikōrero | Apache 2.0 | Ko tātau GPU |
| MadLAD-400 3B | Whakamāoritanga (450+ reo) | Apache 2.0 | Ko tātau GPU |
| Qwen2.5-1.5B (llama.cpp) | Ko te whakarāpopoto, te tātaritanga, te whakanao ihirangi, te kōrerorero RAG | Apache 2.0 | Ko tātau GPU |
| Chatterbox (Resemble AI) | Whakakōrero reo / kupu-ki-whakaaro | MIT | Ko tātau GPU |
| all-MiniLM-L6-v2 | Ko te whakataki i te rapu RAG | Apache 2.0 | Ko tātau GPU |
Ko "STT.ai Enhanced" ko tātau ingoa hua mo Whisper nui-v3 e haere ana i runga i tātau GPU. Ehara i te tauira ā-kaupapa, he pūmautanga, he taumahatanga rānei - ko te Whisper te wāhi whakamātautau nui rawa, tino tika hoki. He tino tika ake i te Whisper nui-v3-turbo pūmau, ā, tata ki te 1.5x pōturi i runga i tētahi mahi whakahuahua, i inetia i runga i tātau ake pūrere.
Kāore e kī mai mātou i te OpenAI, Anthropic, Google Cloud, tētahi atu LLM API tuatoru rānei mō ētahi āhuatanga. Ka haere ia tauira i runga i ngā pūrere e pupuri ana mātou, e whakahaere ana hoki. Ko te ratonga AI waho anake e whakamahia ana e tātau ko translateapi.ai (Muddy Holdings hoki) mō te whakamāoritanga o ngā aho UI - kāore i pā ki ōna ihirangi i whakamāoritia.
He pēhea te whakaatu mai e tātau i te whakatipuranga AI
- Whakawa i ngā pātū HTML: tae atu ki <meta name=\
- Ko ngā kawenga kupu (TXT, SRT, VTT, JSON, CSV, DOCX, PDF): Kei roto i te raina whakahau 'AI-mātautau' i runga i ia pūranga.
- Whakapānga reo whakairo / TTS: Kei roto i ngā pūkete WAV tētahi tohu 'pūoro-whakaaro' i roto i ngā metadata me tētahi mōhiohio mārama i runga i te pātaka whakataki. Ko te whakakāhoretanga pūoro i runga i te mahere huarahi.
- Ka urupare te API: Kei roto i te _ai_generated: āpure pono i ia urupare JSON kei roto i ngā ihirangi i whakamāoritia.
Ko te raraunga whakaakoranga
- Ko ngā tauira taketake (Whisper, MadLAD, Qwen, etc.) e whakaakona ana i mua mai i ō rātou kaiwhakaputa. Ka whakamahia e tātau i te wā e tukuna ana.
- Kāore e whakamahia ana oe ki te whakaako i ngā tauira pūtake.
- Mēnā e tika ana e koe tētahi wāhanga o te whakahuatuhi (te tohuāhua o te rārangi), te tohu rānei he hē (te tohuāhua o te tohu), ā, kua kōwhiria e koe ki /privacy-settings/ ("Model Training" toggle, default off), ka taea te kohikohi i te takirua (tuhi hē → kupu tika) hei whakaako i tētahi tauira whakakore kupu i muri ake. Ko te takirua kupu anake - kāore i te whakawātea i te oro - ka whakamahia e tēnei ara. Kāore he tauira pēnei i tēnei rā: ka rokirokitia anake ngā takirua, ā, kāore anō kia whakaakona e tātau i tēnei wā.
- I te wā kotahi, ko te Whakawhiwhinga i ngā whakarerekētanga me te oro ki te Voice Lab toggle (i /privacy-settings/ hoki, kāore anō te pūnoa) e kōwhiri ana kia whai wāhi ki te oro i te oro o ngā wāhanga e tika ana, e hono ana ki te kupu whakarerekētia, ki tātau pūkete raraunga Voice Lab i raro i te CC-BY-SA-4.0. He takitahi ngā tohenga e rua — ka taea e koe te tuku, e rua rānei, kāore rānei.
- Ka whakakoretia te oro i whakarewaina e koe i roto i ngā wā 24 mā tētahi mahi whakarurutanga i ia wā, me ngā whakarerekētanga e rua e pā ana ki te wā e kōwhiri ana e koe. Mēnā kua kōwhiria e koe ki "Tērā ngā whakarerekētanga + oro ki te Voice Lab" i /privacy-settings/, ka whakakōkītia te oro tae noa ki te 90 rā kia taea ai e te Voice Lab te whakawātea i te wāhanga whakarerekētia; i te wā kotahi (i muri rānei i te 90 rā, i muri rānei), ka whakakoretia te oro pūtake. Mēnā kua tāurutia e koe, ā, ka whakaurua te oro o tōtou whakataki ki tētahi pātahitanga tiritiri, ka pupuritia taua oro kia noho tonu te pātahitanga (ki te 30 rā i runga i tētahi pūkete wātea; 30 rā, 90 rā rānei, kāore rānei i te wāteatanga me te Forever i runga i tētahi kaupapa utu) ā, ka whakakoretia i roto i ngā wā 24 i muri i te whakamutua o te pātahitanga. Kāore te oro tiritiri i whakamahia mō te whakaakoranga. Ko te kupu whakahuatuhi e whai ana i te kaupapahere tiaki i tōtou kaupapahere (24h ā-whakaaro, 30 rā wātea, Pro+ tūturu).
Te tika me ngā hapa
Kāore te whakamāoritanga AI i te tino pai. He rerekē ngā mokatere hapa kupu e ai ki te whakahua o te kaikōrero, te āhuatanga oro, te reo, me te kupu rohe. Mō te whakamahinga hira (ture, rongoā, mahi whakahaere) whakamātau ki te oro taketake. Ko tātau tohu WER tūmatanui mō ia tauira i te /models/.
Ka whakamātautau te whakarongo
Mō te EU AI Act, GDPR, ētahi atu pātai whakarite rānei: hello@stt.ai Ka whakamahia rānei te Tātai āhua.
Whakahōutanga whakamutunga: 2026-08-05.