Frequently Asked Questions about the Speechly SLU platform and API.
Please post a question at our Github discussions page or for a more concrete technical problem, please file an issue.
Try to be as specific as you can. Describe what you are trying to do, how you do it, and what errors (if any) you are getting. We are happy to help you forward, and you will receive an answer much quicker if your question is formulated in a way so that we can easily understand what your problem might be.
Also, if there’s no answer to your question here, you might find one on the Github discussions page.
What are SLU rules?
The SLU rules define the SLU model. The model is configured through sample utterances that are annotated using our custom syntax.
What is an utterance?
An utterance is something that the end-user says. It can consist of one or more segments. See SLU Examples.
How many SLU rules/utterances do I need to train my models with?
It depends fully on the complexity of the model. If your model is only trained to turn the lights on and off, probably 50 lines of utterances is already a pretty good amount. For any more complex application, the number of sample utterances should rather be in thousands.
What is application ID and how do I use it?
An application ID is needed to build an access token for the Speechly API. You can find your application ID in the Speechly Dashboard. The application ID should not be shared publicly, because it allows others to access your model. If you fear your application ID has leaked, you can either delete the application or ask us to revoke access to it.
How do I test my model in practice?
You can test your model in the Speechly Preview.
How long does it take to the train the model?
It depends heavily on configuration. It shouldn’t take more than 5 minutes with most setups, but if you have thousands of lines of rules, it can take longer. You will get a notification once the training is completed.
How do I connect my client to the model I’ve built?
The Speechly API can be called from whatever device or platform. To do this, please review the API Reference to see how to call the API and receive responses. We have also created ready-made libraries / SDKs that help you connect clients on the most popular platforms to the API. You can find documentation for client libraries here.
How can I evaluate the performance of my model?
We will launch our more advanced analytics features shortly, but in the meanwhile, you can test your model in the Speechly Preview.
How do I delete my data?
You can delete all user data from the Admin Dashboard. If you want a verification that all data is deleted, you can contact firstname.lastname@example.org
How is my data protected?
How can I integrate Speechly to my app?
The easiest way to start developing on Speechly is to use our client libraries
How can I integrate Speechly to my app built on another platform?
You can either use our APIs or contact us.
What is device ID and why do I need that?
The device ID is a unique identifier for the end-user device. It is used in our API to differentiate between the different devices that use the model. The application ID is used to adapt the speech recognition to the particular acoustic properties of the microphone on the device, the speaker, and the usage context of the application. Keeping this ID persistent and distinct for each user/device will significantly improve the accuracy of the voice user experience.
The microphone doesn’t work on iOS and Mobile Safari?
iOS doesn’t allow asking for microphone permissions on the page load, but rather, after user interaction. While some browsers might support other patterns, this is the best practice on all browsers, because the user should explicitly initiate the action that requires a microphone. If the permissions are asked on the page load, the user might not understand why it is needed, and may simply deny access.
Make sure your application’s main thread is not blocked by moving all resource heavy computation to other threads by using Web Workers.
Why are my intents and entities not recognized correctly?
The Speechly SLU model is based on machine learning. If your intents and entities are not recognized correctly, we recommend you to add more training rules (i.e., example utterances) to the configuration that contain those intents and entities. Then try again.
I’m having a weird issue and can’t find the answer from the documentation, or the documentation is not correct.
Please post a question at our Github discussions page or for a more concrete technical problem, please file an issue. Try to be as specific as you can, describe what you are trying to do, how you do it, and what errors (if any) you are getting. We are happy to help you forward, and you will receive an answer much quicker if your question is formulated in a way that we can easily address it.
Why do I get “Mic consent denied” error in the Preview? Why doesn’t my microphone work in the Preview?
Once you start the Preview, you’ll have to click on Tap to start and give permissions for your browser to access the microphone. Here’s a short clip that shows the process:
Tap to start
If you have denied access to the microphone and don’t get the dialog for microphone permissions, see these browser-specific instructions:
Last updated by Hannes Heikinheimo on October 17, 2022 at 10:13 +0300
Found an error on our documentation? Please file an issue or make a pull request