Multilingual Speech Data Collection & Annotation
Budget / SalaryHourly project
TypeFreelance project
LocationRemote
Posted47 minutes ago
I’m building a new speech-to-text and voice-assistant pipeline and need a partner who can take charge of the entire speech data workflow—both gathering fresh recordings and producing clean, fully labeled transcripts.
Scope of work
You’ll recruit or source speakers in several languages, guide them through short scripted prompts and natural-dialogue tasks, record the sessions in quiet environments, then transcribe and annotate every utterance. Annotations must include accurate verbatim text, speaker ID, language tag, and start–end timestamps so the files plug straight into my model-training stack.
Deliverables
• High-quality audio files (WAV or FLAC, 16-kHz/16-bit or higher)
• Corresponding annotation files (JSON, CSV, or TextGrid—let’s agree on a format)
• A brief summary report outlining speaker demographics, recording conditions, and hours per language
Acceptance criteria
Recordings meet target sampling rate with no clipping or background noise spikes above –30 dB, transcripts are time-aligned to within ±100 ms, and file naming stays consistent across languages.
If you already have access to multilingual participants or an existing annotation team, let me know; that will speed things up. I’m ready to start as soon as we confirm languages, volume, and timeline.
Scope of work
You’ll recruit or source speakers in several languages, guide them through short scripted prompts and natural-dialogue tasks, record the sessions in quiet environments, then transcribe and annotate every utterance. Annotations must include accurate verbatim text, speaker ID, language tag, and start–end timestamps so the files plug straight into my model-training stack.
Deliverables
• High-quality audio files (WAV or FLAC, 16-kHz/16-bit or higher)
• Corresponding annotation files (JSON, CSV, or TextGrid—let’s agree on a format)
• A brief summary report outlining speaker demographics, recording conditions, and hours per language
Acceptance criteria
Recordings meet target sampling rate with no clipping or background noise spikes above –30 dB, transcripts are time-aligned to within ±100 ms, and file naming stays consistent across languages.
If you already have access to multilingual participants or an existing annotation team, let me know; that will speed things up. I’m ready to start as soon as we confirm languages, volume, and timeline.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.