- Input using speech-to-text
- Fulfillment using Algolia
- Output using speech synthesis
The speech-to-text layer - input
You must have a speech-to-text layer to convert your users’ speech into something Algolia understands (Algolia can’t process non-textual searches). You can add a speech-to-text layer in two ways:- Using the Chrome browser, iOS or Android native apps, or a voice platform tool like Alexa or Google Assistant with speech-to-text built-in.
- Using a third-party service. You send user speech to the service. When you receive it back, you then send it to Algolia as a search query. Some services include:
Algolia: fulfillment
In the fulfillment step, you take user queries and find the results in your Algolia index. You present relevant content to users at the end of this process. There are two parts to the Algolia fulfillment:- Query time settings
- Index configuration
Query time settings
The query time settings improve search results during query time. For instance, selecting a language for Algolia then allows you to set certain features like ignoring “noise” words that users could enter in their search query. If you choose English as the language, and you turn on the stop words feature, the search engine ignores words like ‘a’ and ‘an’ as they’re not relevant to the search query. This gives more exact search results.- Set
removeStopWords
and ensure to select a supported language. For example,en
for English. This setting removes stop words like “a”, “an”, or “the” before running the search query. - Send the entire query string along as
optionalWords
. Speech often has words that aren’t in any of your records. With this setting, records don’t need to match all the words. Records matching more words rank higher. For example, in the spoken query “Show me all blue dresses”, only “blue dresses” may yield results for a clothing store: the other words should be optional. - Set
ignorePlurals
totrue
and ensure to select a supported language. For example,en
for English. This setting marks words like “car” and “cars” as matching terms. - Apply
analyticsTags
to the query, including voice queries. You can activate these settings using thenaturalLanguages
parameter. These settings work well together when the query format is in natural language instead of keywords, for example, when your user performs a voice search.
Index configuration
Similarly, you can apply some rules related to your index. These rules are dynamic and apply depending on what users type in the search query. Detecting user intent can help dynamically change the search results.Speech synthesis: output
Not all voice platforms need speech synthesis or text-to-speech. For example, a site that shows search results may be enough. If your voice platform does need speech synthesis, your options are:- A built-in system such as Alexa or Google Assistant.
- A third-party system. Most modern browsers support speech synthesis through the SpeechSynthesis API. If you want a wider choice of voices, you have Azure Cognitive Services or Amazon Web Services’ Polly.
Prepare your project
To use InstantSearch Android, you need an Algolia account. You can create a new account, or use the following credentials:- Application ID:
latency
- Search API Key:
1f6fd3a6fb973cb08419fe7d288fa4db
- Index name:
instant_search
Create a new Android project
Start by creating a new Android project. Open Android Studio, and select Create New Project.

API 21
as minimum SDK version. Click Finish.


Add project dependencies
In yourbuild.gradle
file at the project level, add mavenCentral()
as a dependency repository in the repositories
blocks:
groovy
build.gradle
file under app
module, add the following in the dependencies
block:
groovy
AndroidManifest.xml
must include the following permissions:
XML
build.gradle
:
groovy
plugins
block to your app’s build.gradle
:
gradle
Create a basic search experience
Start by creating a classic search interface with search bar and results list. First, create a layout file calledlist_item.xml
under res/layout/
.
Add a TextView
to display a hit:
XML
activity_main.xml
to have a minimal set of components for a basic search experience:
SearchView
: view for textual query inputRecyclerView
: the list of search results
XML
Product
to hold search results:
Kotlin
ProductAdapter
to render search results:
Kotlin
MainActivity
use the SearchBox
and Hits
widgets:
Kotlin

Create a voice search experience
This is a two-step process:- Prepare the project for voice input and speech recognition.
- Add a button on the right of the search bar that triggers the voice input.
Setup audio permission
AddRECORD_AUDIO
permission to AndroidManifest.xml
file:
XML
Add voice input UI
Changeactivity_main.xml
to add the new voice input button to the existing UI.
The updated layout file should look as follows:
XML
Add voice input logic
Add a basic implementation of VoiceOverlay: In the click listener, check if you have the audio permission and show the appropriate dialog:Kotlin
VoiceSpeechRecognizer.ResultsListener
Kotlin
MainActivity
should look as follows:
Kotlin

VoiceOverlay
should appear when you tap the voice input button.
At the first launch, it asks for the permissions mentioned in the setup audio permission section.




VoiceOverlay
implementation on the git repository.
Conclusion
With Algolia’s libraries, you can build a voice search experience in less than a hundred lines of code. You can customize your search experience and make it unique by modifyingInstantSearch
components, as well as the VoiceOverlay
components.