Speech is Cheap offers a speech-to-text service. Also known as automatic speech recognition (“ASR”) as well as a transcription service. The service is programmatically accessible via an API.

Main Features

• Fast: the speedup factor is around 100×, meaning that it takes about one minute to transcribe 100 minutes of audio
• Precise: average word error rate (“WER”) is below 8% for English, per the Hugging Face Open ASR Leaderboard
• Inexpensive: at least half as expensive as the nearest competitor
• Multilingual: supports 100+ languages; see the Supported Languages documentation for a list of all languages

Pricing Details

When using the service, the users pay for each minute of transcribed audio, rounded up to the nearest minute. Users of the pay-as-you-go tier pay a flat rate of $0.002 per minute of transcribed audio. That comes out to $0.12 per hour of transcribed audio. Or $0.000033 per second of transcribed audio.

In contrast, monthly subscribers get 21,600 minutes of audio transcriptions per month. When used in its entirety, the pricing comes out to just $0.000926 per minute of transcribed audio. That’s $0.05556 per hour of transcribed audio. Or just below $0.000016 per second of transcribed audio. Any minute over the monthly allowance is billed at the same discounted rate of $0.000926 per minute (or $0.05556 per hour of transcribed audio). That is, overage usage is billed at the same rate as the monthly allowance.

For subscribers, the monthly allowance is reset at the beginning of each month. In return, they are offered a state of the art transcription service (“SOTA”) at the lowest price in the industry. It is possible to switch between the pricing tiers at any time. This way, users may first test-drive the service using the pay-as-you-go tier, and if they find it to be to their liking, they may switch to the monthly subscription in order to take advantage of even lower prices.

Add-ons

It is possible to customize the transcriptions using add-ons. Both tiers can use the additional add-ons on a per-usage basis. Here are the add-ons available:

• Parse Speakers: (AKA speaker diarization) Parses individual speakers based on their unique voice signatures. The add-on costs $0.002 per minute for PAYG users and $0.001 per minute for subscribers. That’s 0.12 per hour for PAYG users and 0.06 per hour for subscribers.
• Parse Words: (AKA word timestamps) Provides per-word timestamps for $0.001 per minute for PAYG users and $0.0005 per minute for subscribers. That’s 0.06 per hour for PAYG users and 0.03 per hour for subscribers.
• Label Audio: Labels audio segments by different categories like speech, music, silence, and so on for $0.0002 per minute for PAYG users and $0.0001 per minute for subscribers. That’s 0.012 per hour for PAYG users and 0.006 per hour for subscribers.

The service is under active development and more exciting and useful add-ons are coming soon!