Audio
Required
Type: File
Uploads the speech recording to transcribe. The pipeline converts it to a mono 16 kHz waveform, but a clear recording with limited background noise will still produce the most reliable transcript.
India600M parameters22 languagesCTC + RNNT
Explore IndicConformer 600M Multilingual
Transcribe speech in any of India’s 22 scheduled languages using the model’s documented waveform, language-code, and decoder inputs. IndicConformer 600M Multilingual is developed by AI4Bharat. Adjust the available settings to shape the result for your task.
Model
ai4bharat/indic-conformer-600m-multilingual
Documentation
IndicConformer 600M Multilingual field reference
Use this reference to understand how every field in the playground changes the request sent to ai4bharat/indic-conformer-600m-multilingual.
lang
Required
Type: String
Default: hi
Values: Assamese, Bengali, Bodo, Dogri, Gujarati, Hindi, Kannada, Konkani, Kashmiri, Maithili, Malayalam, Manipuri, Marathi, Nepali, Odia, Punjabi, Sanskrit, Santali, Sindhi, Tamil, Telugu, Urdu
Identifies the spoken language using the model’s required language code. Choose the language actually heard in the recording; the model does not use this field as an automatic-detection hint, and an incorrect code can substantially reduce transcription accuracy.
decoding
Required
Type: String
Default: ctc
Values: CTC, RNNT
Selects the output decoder used to turn acoustic model predictions into text. CTC is simpler and generally faster, while RNNT models token history during decoding and may handle conversational timing differently; compare both for the target language and audio conditions.