What do you need help with?
Question
Issue area
Speech Recognition / STT
Impact
Blocker - I cannot continue
Character ID
c2e8f824-6e58-11f1-9a04-42010a7be02e
Session ID
290d8a24c1e7d27ede2307e21454d7ce
Where does it happen?
API / Backend
Language
Spanish
Voice provider
Azure voices
Voice name
Marcello Azure voices
Short summary
Speaking English → fully transcribed and understood correctly, full conversation works. Speaking Spanish → only 1–2 unrelated english words
User input
Any word in spanish
What happened?
When sending recorded audio as a file (via the file field in POST /character/getResponse, WAV mono 16kHz) with Spanish speech, only 1–2 unrelated English words are transcribed/received by the character — as if speech recognition defaults to English regardless of the configured language. Sending the same request with English speech in the audio file works correctly and the full sentence is transcribed and understood. I have not tested this with the userText (text-only) input, since the issue only occurs with the audio-file path.
What did you expect?
Since Spanish (es-ES) is configured as one of the character’s languages under “Language and Speech,” I expected the full Spanish sentence to be recognized and understood correctly, the same way English is — either via automatic language detection or by respecting the languages configured for the character.
Steps to reproduce
Configure a character in the Convai Playground with 4 languages under “Language and Speech,” including Spanish (es-ES) with its own assigned voice.
From a custom web app, record microphone audio and encode it as WAV (mono, 16kHz).
Send a POST request to https://api.convai.com/character/getResponse with the audio in the file field, plus charID, sessionID (-1 for a new session), and voiceResponse=True.
Speak in English into the recording → the character responds correctly, full sentence understood.
Repeat the exact same steps speaking in Spanish instead → only 1–2 unrelated English words are received/transcribed; the rest of the sentence is lost, and the character’s reply doesn’t make sense given what was actually said.
I checked the relevant documentation, resources, and similar forum posts before posting.
on
I confirm that I did not include API keys, tokens, passwords, secrets, private character data, private conversation logs, or customer data.
on