Bed Word: Automated Interview Transcription via Deepgram

Bed Word requires a Deepgram API Key to use. To create an account and get a key, follow the instructions here. Bed Word uses Deepgram to transcribe audio then formats the result into output files (including TextGrids). You also have the option to directly send this automated transcription to DARLA's Alignment and Vowel Extraction system.

Per Deepgram's current pricing, one hour of audio transcription will cost $0.25 to transcribe with the Nova model. Please note if you use Bed Word, your data will be sent to a third-party outside of the DARLA system. (Deepgram privacy notice).

Click here at any time to cancel or start over.

Deepgram API Key:
You can access your API Key from your Deepgram account.

Deepgram transcription tier: Nova (default) Enhanced Base
As per 8/27/23, Nova, Enhanced, and Base cost $0.25, $0.87, and $0.75 per audio hour, respectively.
Nova is cheaper and more accurate than Base but is not available in every language. Enhanced is the most accurate and most expensive tier.
For the most up-to-date pricing, please view your Deepgram account's billing page.

Audio file:

Transcription language:
Please see the list of supported languages and tiers here . Please note that
not every language supports every tier; we will automatically select the cheapest available tier for the language if you choose an unsupported tier.

Remove interviewer transcription? Yes No
By default, Deepgram will transcribe everything and will not identify speakers.
If 'Yes' is selected, Deepgram will try to identify two separate speakers in the audio, and only keep the interviewee (assumed to be the person who speaks more).

Include punctuation in output? Yes No

Send results through DARLA's Alignment and Vowel Extraction system after? Yes No
This will happen after your transcription and you'll automatically be emailed the results.
As DARLA is based off the CMU Dictionary, it can only accept English inputs.

Filter out stop-words? Yes No
This is the list of stop-words we remove. (Link opens in a new tab.)

Filter out unstressed vowels? Yes No

Filter out vowels whose F1 or F2 bandwidths are over 300? Yes No

Voice type: Low High

Your e-mail address:
We will not store or distribute your address.