Speech Note - speech to text on Sailfish OS: how good is it?
A quick look at Speech Note, an application providing speech to text functionality: how to install it, and testing its accuracy, both in French and English.
Speech Note?
Introduction
As the name suggests, Speech Note provides speech to text features, which you can
use e.g. when composing a text, an email or a note.
It is comparable to Apple's Siri, or Google Voice Typing.
It works with Coqui STT (Speech-To-Text), a fork of Mozilla's DeepSpeech.
All voice processing is entirely done locally on the device, meaning there's nothing going to the Cloud - Internet connection is only required to download the acoustic models, when doing the first configuration (see below).
Installation
Speech Note is only available on Openrepos.
As usual, you can install it either by downloading the RPM and installing it, or using Storeman.
Configuration
Configuring Speech Note is simple: open it, and go to Settings in the pulley
menu.
This will give you access to:
- Languages: choose which languages you want to use
- Listening mode: how Speech Note behaves with regards to speech recognition
- One sentence: listen until the first sentence is recognised
- Press and hold: listen is on when you press and hold, and off when you release
- Always-on: always listens
- Location of language files: where to store them
Once configured, you have to go in SFOS Settings > System > Text input >
Keyboards, and select the languages you want to add (you'll see an icon which
looks like someone talking):
Once enabled, you will see additional options in the keyboard layouts (either when swiping from left to right/right to left, or when holding the space bar):
In my case, I am using English and French ("Français").
Test
The below tests have been done with SFOS' Messages application, but Speech Note works exactly the same way with any other app, as soon as you have text to enter.
French
I first tested French, since this is my native tongue.
I started with a very simple sentence: "Ceci est juste un test" (i.e. "this is
just a test).
As you can see below, no problem at all:
Then I continued the sentence, making it a bit longer: "pour voir si le clavier à reconnaissance vocale fonctionne bien" ("to see if the voice recognition keyboard works well).
As you can see, Speech Note made a small mistake, confusing "si" (if) with "c'est" (it is).
I then finished with "correction, pour voir si le clavier à reconnaissance vocale fonctionne bien"
Speech Note confused "reconnaissance" (recognition) with "connaissance" (knowledge), and "vocale" (voice) with "locale" (local).
In both cases, not a big deal, as I am a fast talker, so retrying a bit slower gave much better results.
English
I was curious to see how good the recognition would be with a non-native Engish speaker: I think my English skills are OK, but I still have a French accent, so let's see how Speech Note deals with that :) (I'm not adding screenshots, but just the results).
Don't pay attention to the meaning or the context of the sentences, I just tried with some random sentences I had in mind:
- "This is just a test" => "this is just a test"
- "to see how good this software is" => "to see how good this sort were is"
- "can you call me back please?" => "can you call me back please"
- "I don't think we already met" => "i don't think we already met"
Wrap-Up
I am pleasantly surprised by how good Speech Notes is: sure, it's not perfect,
and it does not handle punctuation (e.g. you can't pronounce "comma" and have it
do a , for you), but I still find it useful, especially since it's 100%
offline.
I don't use it as a replacement of my keyboard typing, but rather to complete it.
Tags: SailfishOS