Choosing Languages and Voices

An agent has two speech settings that people mix up: the language it listens in (speech-to-text, STT) and the voice it speaks with (text-to-speech, TTS). They are set separately, and neither list is fixed by Wetel.

The source of truth is Google’s, not this page

Wetel does not keep its own list of supported languages or voices. The speech engines are Google’s, and Google adds and retires languages and voices on its own schedule, so any list copied here would go stale.

• Speech-to-text languages: Google’s list.
• Text-to-speech voices: Google’s voice list.
• Both links are also returned by the agentConfigOptions query as sttLanguageCatalogUrl and ttsVoiceCatalogUrl.

If this site and Google’s lists disagree, trust Google’s, and please tell your Wetel contact so we can fix the page.

Which setting do I need?

Your situation	Set this
Users speak one language	sttLanguageCode to that language, sttAlternativeLanguageCodes to []
Users mix two languages (for example English and Malay)	Leave the defaults: the primary follows ttsLang, alternatives are en-US and ms-MY
Users speak Chinese	Make a Chinese code the primary (cmn-Hans-CN), and list up to two others as alternatives
You are not sure what users speak	Use the Gemini speech engine, which detects the language itself (see below)
You want a specific persona’s sound	Pick a specific voice, and keep ttsLang and voiceTier consistent with it
You only want a natural-sounding default	Pick a voice tier and a language, and leave the voice name alone

Speech-to-text: primary and alternatives

• Primary (sttLanguageCode): the language the recognizer expects. It has by far the biggest effect on accuracy. If you leave it empty, the agent’s ttsLang is used.
• Alternatives (sttAlternativeLanguageCodes): extra languages the recognizer may switch to. At most 3, no duplicates, and the primary must not be repeated. Google silently ignores the whole list when it holds more than three, so Wetel rejects it up front.
• Default alternatives: when you set nothing, Wetel uses en-US and ms-MY. An explicit empty list ([]) means “no alternatives”, which is different from not setting it.
• Never put a secondary language as the primary. A bilingual agent should keep its main language as the primary and add the other as an alternative.
• Chinese is not in the defaults on purpose. Google tended to label accented English as Mandarin, which produced empty or garbled transcripts. If you need Chinese, set it as the primary, or list cmn-Hans-CN or yue-Hant-HK explicitly.
• Not supported at any setting: Hokkien and Hakka.
• Mixing languages in one sentence is not handled well: the recognizer picks one dominant language and usually loses the other’s words.

The two speech engines behave differently. Cloud Speech-to-Text uses the primary and alternatives above. The Gemini engine works out the language itself and ignores both settings. See the /stt reference for which engine your environment uses by default and how to choose one per request.

Text-to-speech: tier, language and voice

A Google voice name has the form <language>-<tier>-<name>, for example en-US-Chirp3-HD-Kore. The name already contains the language and the tier, so three settings have to agree:

• ttsLang: the language, such as en-US or ms-MY.
• voiceTier: STANDARD, WAVENET, NEURAL2, CHIRP3_HD or STUDIO.
• ttsVoice: the full voice name.

If they disagree (for example a Neural2 voice name with voiceTier set to STANDARD), the request may be rejected or the voice may be ignored, so don’t rely on either outcome. Set all three from the same row of Google’s voice list.

Tier	Use it for	Cost
STANDARD	Cheapest, clearly synthetic	Lowest
WAVENET	Better quality, still budget	Low
NEURAL2	Natural, a safe default	Medium
CHIRP3_HD	The most natural for live conversation	About 2x NEURAL2
STUDIO	Narration and recorded content. Not for real-time chat	About 10x NEURAL2

Use agentConfigOptions for a short, curated voice list, but remember it is a starting point and not the full catalog.

Worked examples

One language, no guessing:

mutation {
  updateAgent(
    id: 12
    input: {
      ttsLang: "ms-MY"
      sttLanguageCode: "ms-MY"
      sttAlternativeLanguageCodes: []
    }
  ) {
    id
  }
}

Mandarin primary with English as a backup:

mutation {
  updateAgent(
    id: 12
    input: {
      ttsLang: "cmn-CN"
      sttLanguageCode: "cmn-Hans-CN"
      sttAlternativeLanguageCodes: ["en-US"]
    }
  ) {
    id
  }
}

Troubleshooting

• Empty transcript, or the wrong script: the primary language is probably not what the speaker used. Fix the primary first; adding alternatives helps less.
• confidence of 0: it means “not reported”, not “low”. See the /stt reference.
• A voice sounds wrong or is rejected: check the voice name against Google’s list, and check that ttsLang and voiceTier match it.