7 Questions to Help you Choose the Right Speech-to-Text API

integrating an api to your project is a big commitment so here are seven questions to answer to help you select the best speech to text api number one

how accurate is this api accuracy of automatic speech recognition systems or asrs for short is calculated using a metric called word error rate or w e r word error rate is the number of errors that are made during the transcription divided by the total number of words before selecting an api and integrating it in your system make sure you go ahead and check their benchmark results with respect to wer you can find an example of a benchmark report in the description below it will be titled 2021 benchmark report by assembly ai in that report we are comparing assembly ai's api's performance with some other company's api's performances if you would like to get a custom benchmark report that is built on the data set that you have and for your specific use case you can ask assembly ai to create one for you so in these benchmark reports we are using the word error rate metric to compare different models to each other but word error rate even though it's a very useful single number metric has some shortcomings of its own here's an example to illustrate this issue so let's say what was spoken was that my name is paul and i am an engineer model one predicts my name is ball and i am an engineer whereas model 2 predicts my name is paul and i'm an engineer if we only use the word error rate the first model would get a word error rate of 11. 11 whereas model 2 22. 22 as you can see even though model 2 predicts a more accurate result model 1 gets a lower error rate so a higher accuracy this happens because word error rate does not care about the context and it only focuses on the amount of mistakes that were made unfortunately there is not yet a better way to compare speech to text models but keeping these shortcomings in mind will help you make a healthier decision on top of the benchmark report if you'd like you can also use another tool called diff checker div checker basically helps you look at two transcriptions at the same time and see what their differences are by doing this you can manually check two different transcriptions and see which one gives better results for your specific use case together wers a word error rate and a div checker can be a very strong approach to choosing the best speech to text api for yourself the

second question you should ask is what additional features does this api offer depending on your project you might need more than just a plain transcription of your audio files choosing an api that offers additional features could be a good investment for the future as your project evolves some commonly offered features by apis are topic detection chapterization speaker diarization sentiment analysis and more these features will help you get more value out of your audio files and will lead to a wider variety of use cases that you can build number three is the pricing

clear before starting to use an api make sure the pricing is clearly stated on their website this includes the base pricing pricing of any extra features and any discounts that you might get if you are a high volume user watch out for hidden extra costs that might substantially increase the price next is the

documentation comprehensive and clear the documentation of the api will be your main source of reference while you are integrating the api in your project but also in the future while you're maintaining your project check to make sure that the api has easily accessible structured and understandable documentation the fifth question that

you should ask is what kind of support can i expect even if the documentation of the api is spot on for your specific case you might need personalized help in that help from a developer from the company or from the customer service make sure you select the company that has support available for you in a timely manner to avoid a lot of headache down the road number six

how secure is my data with this api data security is one of the main considerations when integrating a new api to your tech stack by making sure that the api you choose is not going to use your data in any other way than it was intended you will make sure your data is safe some questions you can ask to ensure the protection of your data are does the api keep a copy of my audio or video files in order to improve its model does the api keep a copy of my transcription files keeping a copy of your data to improve the company's own api's performance might seem like an innocent thing to do for you but if you have any sensitive data or any personal information of anyone it might be problematic to include that data to train a model because then the learnings of this model will include that personalized or sensitive information some other questions you can ask on top of this are if the api does keep a copy can i request that it permanently delete my audio or video or transcription files at any time how quickly will my request be met and finally does the api monetize my data before you start working with an api make sure you research the answers to these questions and if you cannot find all of the answers on their website on other resources you can try getting in contact with their customer support that could also be a good way to check how fast the customer support will get back to you and the last question that

you should ask is innovation a priority for this company speech to text technology is still evolving new approaches are being proposed and breakthroughs are made continuously to make sure that your project is future proof it is a good idea to work with an api that evolves over time with the latest innovations and keeping up to date most of the time will require a on staff research team checking the updates and the changelog of a company is a good way of making sure that a company does not only claim that they prioritize innovation but they do actually innovate and by checking the updates and the change log you can also see how well the api and its extra features are maintained there are many things to consider when you're choosing a special text api and it can easily get overwhelming but remember that doing your research to make sure that you're selecting the best speech to text api for you that is well supported well maintained and well documented will set you up for long-term success if you would like to give assembly ai a try you can use the link in the description to get your free api token to get started with it and experiment with it but for now thanks for watching and i will see you in the next video