OutStanding DataTech-Audio Collection

OSDT brings a wealth of experience in offering multilingual audio collection and annotation services for AI applications. Our team is adept at capturing top-notch audio data in a wide range of languages, guaranteeing accuracy and uniformity. By adhering to strict guidelines and implementing stringent quality control measures, we deliver precise annotations in multiple languages. Our expertise in navigating diverse linguistic nuances and cultural contexts allows us to cater to the unique needs of each project.

Resources and collection tools

Over 20 cooperative recording studios in different cities.

15 resource manager responsible for global recruitment.

Self-developed intergrated data collection platform applied worldwide.

Wake-up words collection key points

Quality Control

Acceptance standard is a key link in the project implementation process. By setting a clear standard the team can improve the operation efficiency.

Site Requirements

Except for specific projects that need to be recorded in a professional recording studio, most of wake-up words collection generally need to be performed in a relatively small and quiet room.

Features of different scenes

If the model being trained needs to collect audio with a certain background noise, it needs resources to find a suitable environment for recording.

Tool usage process

For requirements like the speed of speaking, blank space ahead or after the reading, a restricted operation process of each step of the resource on the operation platform will reduce the probability of making mistakes.

Equipment Requiremet

Mobile phones are preferred and seldom of the recording use headphones, otherwise there will be uncontrollable bottom noise or spraying wheat.

Confirmation of rules before execution

Like collected equipment requirements, audio channel and other parameter values, those specific requirements need to be confirmed before excuation.

Audio Collection

Collection Scenes	Metadata	Topics	Device	Software	Output format	Distance
Voice-controlled command	Speaker ID, Name, Age, Gender, Demographic Info, Mother tongue, Script, Time stamp, Device used, Distance	Wake-up word, Control commands	Mobile phone	Collection app	Bit depth, Frequency, Channel, other format,	0.3-3m
Free conversation		By industry field	Laptop, Mobile phone, Microphone	Target application, Collection app		0.5-1m
Single-person recording		Designated script	Laptop, Mobile phone, Microphone	Collection app		0.5-1m
Environmental noise	Date, Scenario	N/A	Mobile phone	Collection app		2-5m
TTS	Professional voice actor with designated tone	Designated script	High-fidelity microphone	Studio		0.3m

**Requirement List**
Factors	Definition
Applied Scenarios	Specific scenarios where this project will be used to make all stakeholders have a better understanding
Workload	Workload for each participant, proportion duration of short or long audio, TAT for the whole project
Script	Provided by client, generated by OSDT, generated by individuals
Language	Mother languages, tongues, accents
Domains	History, Agriculture, Culture, Finance, Holiday and Leisure, Art, Comedy, Drama, Hospitality, Aviation, Entertainment, Gaming, Information and Technology, Banking, Crime and Justice, Health, Insurance, Religion, Study, Retail, Technology, Weather, Politics, Spirituality, Travel..
Resource	Ratio of male to female, race, geographic location, age group
Measurement Units	Total audio duration, effective audio duration, number of items (by Seconds, minutes, hours, sentences, utterances)
Collection Devices	mobile phones, recording studios, laptop, etc. It is needed to have the white list for applicable device (like type and configuration requirements)
Collection Software	Provided by client or by OSDT, web-based link or Application to install
Collection Scenarios	Indoor or outdoor, inside or outside a car, window open or closed, office environment, meeting environment
Collection Rules	Single-person, multi-person conversations, speech speed, tone Repeats, re-starts, ﬁller words, etc., should/shouldn’t all be present, use authentic voice and style.
PII	Do not share any real personally identifiable information during the recording (e.g. real names, addresses, phone numbers, etc.)
Collection Distance	Close-range headphones, close-range earpieces, long-range, etc.
Audio Quality	1. Background noise requirements (SNR, db) 2. Sampling rate (16kHz, 24kHz, 44kHz, 48kHz), the hardware will be the limit, not the software. 3. Bit depth (16 bits, 24 bits, 32 bits) 3. Channels (mono or stereo) 4. File format (wav, mp3...) 5. File name rules
Acceptance	Frequency: by batch or one-time final acceptance. Review ratio: full review or random sampling by the client.

Project Experience