Code-Switching ASR · മലയാളം
Malayalam Data for Code-Switching ASR
Recognising speech that switches between an Indian language and English several times per sentence. In Malayalam, the binding constraint is usually dialect coverage and code-mixing, not raw hours.
- Language
- Malayalam
- Primary metric
- Switch-point accuracy
- Typical volume
- 100-500 hours
Data profile required
- Genuinely code-mixed spontaneous speech
- Per-token language ID labels
- A fixed rule for script of English tokens
What Malayalam adds to the requirement
- One of the most consonant-dense Indian languages; long geminates and clusters raise word error rates sharply
- Very high speech rate compared with other Indian languages, which stresses streaming ASR
- Dialects to cover: Thiruvananthapuram, Kochi (central), Malabar / Kozhikode, Thrissur
- Manglish is standard in urban and professional speech, with heavy English noun and verb insertion.
Metrics to track
- Switch-point accuracy
- Mixed-utterance WER
- Language ID token accuracy
Failure modes
- Concatenating monolingual data and calling it code-mixed
- Leaving script conventions to individual annotators
For Malayalam specifically: Central Kerala news-reading dominates public data. Malabar and southern varieties, and fast conversational speech generally, are missing.
Recommended cohort
Budget higher transcription effort per audio hour for Malayalam than for Hindi; speech rate and morphology make it slower to annotate.
| Dimension | Typical split | Why it matters for Malayalam |
|---|---|---|
| Gender | 50 / 50 | Pitch range differences change acoustic model behaviour; unbalanced cohorts bias recognition |
| Age | 18-25: 30%, 26-40: 40%, 41-60: 30% | Older speakers retain conservative Malayalam forms that younger urban speakers have lost |
| Region | Kerala / Lakshadweep / Puducherry (Mahe) and others | Dialect spread across 5 recognised varieties |
| Education | Mixed, including below-graduate | Prompt-reading fluency correlates with education and skews prosody |
| Condition | Studio / quiet room / field | Match the noise profile of your deployment |
Suggested programme shape
Start with an evaluation set of 100 speakers spread across every Malayalam dialect in scope, collected before training data. Then field 100-500 hours of training data from disjoint speakers.
This ordering is what makes the improvement measurable rather than assumed.
Frequently asked
Is there usable public Malayalam data for code-switching asr?
Central Kerala news-reading dominates public data. Malabar and southern varieties, and fast conversational speech generally, are missing.
How many Malayalam speakers do we need?
300-800 speakers for a training corpus, plus a disjoint evaluation cohort covering each dialect. Speaker count matters more than hours for generalisation.
Can you run this across multiple languages at once?
Yes. Multi-language programmes run to one master specification so per-language results stay comparable.
Scope Malayalam data for code-switching asr
Send the target metric and the languages in scope.