Wake-Word & Command Corpus

[ OVERVIEW ]

Short-form utterance corpora built specifically for voice-interface models: wake words, hotwords, command phrases, and directed-dialog turns. Recordings are captured under controlled conditions with defined acoustic variation (quiet, ambient, distant-field, and adversarial noise). Every utterance includes speaker demographics, environment class, and signal-to-noise metadata. Scoped per wake-word, per language, or across demographic distributions your product team needs to cover.

[ KEY HIGHLIGHTS ]

• Controlled acoustic conditions across quiet, ambient, distant-field, and noisy environments
• Speaker demographic coverage by age, gender, accent, and language
• Signal-to-noise and reverberation metadata per utterance
• Custom wake-word and hotword coverage scoped to your product
• Directed-dialog commands with intent classification labels
• Far-field and adversarial-noise subsets available
• Licensed per-wake-word or as the full voice-interface corpus

[ TECHNICAL SPECIFICATIONS ]

Files

Mono WAV, 16-48 kHz, 16-bit, with environment class per recording

Transcripts

JSON with utterance text, speaker metadata, environment class, SNR

Annotations

Wake-word / hotword / command tag · intent classification · environment label

Licensing

Commercial training rights · per-wake-word or full-corpus · demographic distributions scoped per project