STT/ASR Competitive Analysis

Delivered January 24th, 2020. Contributors: MeliSan and

Part

of four

Part

STT/ASR Competitive Analysis: Google & IBM

Google's Cloud Speech-to-Text API is available in 120 languages as a cloud service, and as an application, while IBM's Watson STT is a cloud service available in 17 languages. The details of each product were added to the attached spreadsheet and a summary of our findings below.
GoogleGoogle's Cloud Speech-to-Text API is available in 120 languages and different variants for free between 0 and 60 minutes and for a price between $0.004 and $0.009 per 15 seconds for speech-to-text recognition above 60 minutes.
While the product was created as a cloud service, it is also available for desktops, laptops, phones, or tablets.
Additionally, it allows users to customize up to 5,000 words, phrases, and other specific business words like years, addresses, currencies, conversions, etc.
IBMIBM's Speech-to-Text program, Watson, is a cloud service available in 17 languages and variations, under three plans: Lite, Standard, and Premium.
Lite: offers between 100 and 500 minutes per month for free without access to customization.
Standard: Charges between $0.02 and $0.015 per minute between one and over 1 million minutes.
Premium: Charges vary based on model combinations and customization.
The product allows language customization based on business needs, as seen in the medical language sample video.

Part

of four

Part

STT/ASR Competitive Analysis: Microsoft & Amzon's Web Services

Microsoft offers Cognitive, an STT service available for devices and cloud in 40 languages from $1 per hour to $2.10 per hour. Amazon's Transcribe provides cloud STT and ASR services in 31 languages from $0.006 per 10 seconds to $6.75 for 90 minutes. The details of our research were added to the attached spreadsheet and a summary of our findings below.
MicrosoftMicrosoft Azure's STT service, Cognitive, provides multiple speech services, including STT in 40 languages and variations on the cloud and on devices. 
The users get 5 hours of STT for free every month on the standard and custom models.
After the 5 hours, the price is $1 per audio hour on Standard, $1.40 per hour on custom model adding $0.0538 per hour per additional customization, and $2.10 per hour per conversation or multichannel audio transcriptions.
The service allows to customize models and languages, as shown in the attached example video.
AmazonAmazon's STT and ASR technology, Amazon Transcribe, is a cloud service that supports 31 languages and variations.
During the first year, users get 60 minutes per month for free, after that, the Standard plan's price starts at $0.006 per 10 seconds and goes up to $4.80 for 200 minutes, depending on the type of audio and length. 
If the user chooses the medical transcriber, the price goes from $1.125 for 15 minutes to $6.75 for 90 minutes, depending on the type of audio.
An example of ASR customization using Amazon Transcribe can be seen in this YouTube video.

Part

of four

Part

STT/ASR Competitive Analysis

Two additional companies that offer Speech-to-Text or Automatic Speech Recognition products are Huawei and Speechmatics. The names of the companies were added to the attached spreadsheet under rows 6 and 7. Find an overview of the products below.
1. HuaweiHuawei Cloud offers Automatic Speech Recognition (ASR).
Their service converts each one-minute audio recording into 4MB text (SST) automatically after uploading the audio file to the platform.
The service has a 95% accuracy, and also allows strong language support.
2. SpeechmaticsSpeechmatics offers different speech recognition services, including ASR and STT, for audio and video.
Their coverage is global, with a wide list of languages supported and transcription results in minutes.
Its error rate is the lowest in comparison with competitors like Google and Kaldi.

Part

of four

Part

STT/ASR Competitive Analysis: Rows 6 and 7

Huawei Cloud is an ASR service available in four languages, from US$549.17 per 20 million calls to US$2,288.19 for 100 million calls. Speechmatics provides SST and ASR services in 74 languages, for $2.4 to $3.6 per hour of audio. The details of our research were added to the attached spreadsheet and a summary of our findings below.
HuaweiHuawei Cloud is a cloud-based speech recognition service that offers ASR in the languages of English, Portuguese, Simplified Chinese, Spanish, and Mandarin.
The users are billed every month based on their use, for a unit price of ¥0.004 per calls per hour. The packaging prices cover a year of service of 20 million API calls for ¥60,000 or US$549.17, 50 million API calls for ¥140,000 or US$1,281.39, and 100 million API calls for ¥250,000 or US$2,288.19.
It allows ASR Customization (ASRC) through deep learning and also the possibility to optimize speech recognition based on business needs and language models.
SpeechmaticsSpeechmatics provides ASR and STT through the cloud and on-premises in 74 languages and variations.
The service is sold in packages, and although the price is customized per client and not available on its website, the market comparisons indicate that their cost per minute decreases based on the package acquired. 
The starting package is of $10,000, and the cost per hour of audio goes from $2.4 to $3.6 approximately.
This tool allows custom dictionary and sounds, as shown in the video example.