Why is it that even though I set the language to Chinese, the speech model still recognizes English text?

[LiqiNew]

private val DEMO_STT_MODEL = SttModel.PARAKEET

            val config = SpeechConfig(
                modelDir = modelDir,
                useNnapi = false,
                sttModel = DEMO_STT_MODEL,
                precision = ModelPrecision.INT8,
                emitPartialTranscriptions = true,
                partialTranscriptionInterval = 0.5f,
                language = "cn",
                languageHints = listOf("en", "cn")
            )

            val p = pipelineFactory(config)

'''Why is it that even though I set the language to Chinese, the speech model still recognizes English text?'''

[ivan-digital]

Fixed in #75.

Parakeet TDT v3 supports 25 European languages, always performs automatic language detection, and cannot honor language or languageHints. It does not support Chinese, and cn is a country code rather than a Chinese language tag.

The SDK now rejects these configurations instead of silently ignoring them. For Chinese recognition, use SttModel.NEMOTRON_MULTILINGUAL with the LiteRT backend and language = "zh-CN" or "zh-TW". The SDK documentation and all translations have also been corrected.

[ivan-digital]

The fix is now available in v0.0.16.

• Maven: audio.soniqo:speech:0.0.16
• Signed demo APK: https://github.com/soniqo/speech-android/releases/latest/download/app-release.apk
• Signed control-demo APK: https://github.com/soniqo/speech-android/releases/latest/download/control-demo-release.apk

The release passed unit/assembly CI, CodeQL, and the full arm64 Android connected suite with zero failures. Existing Parakeet integrations should remove concrete language and non-empty languageHints; use Nemotron with zh-CN or zh-TW for Chinese recognition.