Speech Corpus Recording Consultant — Multi-Speaker, Multi-Mic Array Capture

Job Description We are building a large-scale, multi-speaker conversational speech dataset for a commercial AI/speech-separation use case (diarization and speaker-separation model training). We need an experienced audio/acoustic engineer to design our recording protocol, guide our in-house team through execution, and provide ongoing quality review as we scale data collection across multiple locations. This is not a music recording, mixing, or mastering project. We need raw, unprocessed, scientifically clean multi-channel capture — not a "good sounding" mix. Project Background - We are collecting real, co-located, multi-speaker conversational audio to serve as training/evaluation data for speech AI models (speaker diarization / source separation). The dataset must meet strict technical requirements around synchronization, signal isolation, and lack of post-processing, detailed below. Hard Technical Requirements (Non-Negotiable) - 2+ speakers, physically co-located in the same room — no remote calls, no replayed/re-recorded audio One close-talk microphone per speaker (headset, lapel, or binaural) — raw capture, with no noise suppression, AGC, or beamforming applied At least one far-field microphone (room mic or mic array) capturing the full mixed signal Sample-level synchronization across all channels — either a shared clock/single interface, or verified alignment (e.g., clap/tone sync) across independently clocked devices Natural conversational dynamics — spontaneous turn-taking, interruptions, natural overlapping speech (target overlap ratio ≥10%) Uncompressed PCM WAV output, 16-bit or 24-bit, consistent sample rate across channels (16kHz or 48kHz) Environmental variety — multiple room types/acoustic conditions, varying speaker-to-mic distances (1–4m for far-field), a mix of recording hardware (laptop mics, phone mics, studio mics, lapel mics, arrays) What We Need From You - Protocol design: A detailed, written recording protocol covering equipment selection/setup, mic placement guidelines, sync verification method, session structure, and a pre-session checklist our team can follow without prior audio engineering expertise. Equipment recommendations: Specific mic/interface/recorder models (with budget tiers) suited to close-talk + far-field simultaneous capture, appropriate for repeated use across varied rooms in India. Sync verification method: A concrete, repeatable way for our non-expert team to verify sample-level sync at the start/end of every session. QC criteria & rejection checklist: Clear pass/fail criteria (clipping, bleed thresholds, dropout, sync drift) our team can apply to every recorded session before it's accepted into the dataset. Team training: One or more live/recorded training sessions (video call or in-person if Ahmedabad-based) to walk our operations team through correct setup and common failure modes. Ongoing review (optional/extendable): Periodic review of sample recordings as we scale, with feedback on drift from spec. Ideal Candidate Background - -Experience with multi-channel/array microphone recording (not just stereo music tracking) -Background in speech corpus collection, ASR/voice-AI data collection, broadcast/film production sound (multi-lav + boom sync), or acoustic/bioacoustic research -Comfortable explaining why raw/unprocessed capture matters (i.e., understands downstream ML use, not just "clean audio" from a listening perspective) -Prior work with companies or labs in speech AI, telecom, linguistics, or corpus vendors (e.g., LDC/ELRA-style collection, or Indian voice-AI companies) is a strong plus -India-based or willing to work across IST hours; based in or able to travel to Ahmedabad is a plus but remote consulting is acceptable