speech_to_text 2.7.0

A Flutter plugin that exposes device specific speech to text recognition capability.

Changelog

2.7.0

New

• the example had onDevice true which is a rarely used flag and was causing confusion
• the partialResults option for the SpeechToTextProvider.listen was defaulted to false which is a less common usage and didn't match the default for SpeechToText.listen so it has been changed to true

2.6.0

New

• new parameter onDevice on the SpeechToTextProvider listen method supports forcing offline recognition.
• Added a new Tips section to the README doc to answer some common questions.

Fix

• Android now correctly returns multiple possible matches with confidence
• Android now supports the onDevice flag properly, note that at least some Android devices need an offline language pack installed for each target language to enable offline recognition.

2.5.0

New

• new parameter localeId on the SpeechToTextProvider listen method supports selecting a non default languge for the recognizer when using the provider.

Fix

• A work around for a bug in Android 10(29) that made stop and cancel not work. The plugin now calls destroy on the Android SpeechRecognizer which terminates the listen session immediately. There is a downside to this approach in that it does not always report results that were in process when destroy was called, however it does mean that stop works.
• The example for SpeechToTextProvider, provider_example.dart had an error that made it always use the default language and ignore the drop-down selection. This has been corrected.

2.4.1

Fix

• Updated the version number in the readme file, which I forget to do, every single time.

2.4.0

New

• new parameter sampleRate on the listen method supports some older iOS devices by allowing customization to support the expected hardware sample rate. 44100 works with some older devices.

Fix

• pauseFor now times out closer to the expected time and more reliably. Duplicate partial results were causing it to extend previously.
• finalResult is now reliably true on the last result on iOS when pauseFor or listenFor timeout or stop is called directly. Previously it would only return true when the stop happened almost immediately

2.3.0

New

• new parameter onDevice on the listen method enforces on device recognition for sensitive content
• onSoundLevelChange now supported on iOS
• added compile troubleshooting help to README.md
• SpeechToTextProvider is an alternate and simpler way to interact with the SpeechToText plugin.
• new provider_example.dart example for usage of SpeechToTextProvider.

Fix

• on iOS handles some conflicts with other applications better to keep speech working after calls for example

2.2.0

New

• improved error handling and logging in the iOS implementation
• added general guides for iOS to the README
• moved stress testing out of the main example
• iOS now defaults to using the speaker rather than the receiver for start /stop sounds when no headphones

Fix

• iOS now properly deactivates the audio session when no longer listening
• start and stop sounds on iOS should be more reliable when available

2.1.0

Breaking

• listenFor now calls stop rather than cancel as this seems like more useful behaviour

Fix

• Android no longer stops or cancels the speech recognizer if it has already been shutdown by a timeout or other platform behaviour.
• Android no longer tries to restart the listener when it is already active
• Now properly notifies errors that happen after listening stops due to platform callback rather than client request. See https://github.com/csdcorp/speech_to_text/issues/51

2.0.1

Fix

• Resolves an issue with the Android implementation not handling permission requests properly on apps that didn't use the 1.12.x plugin APIs for registration. The permission dialog would not appear and permission was denied.

2.0.0

Breaking

• Upgraded to New Swift 1.12 plugin structure, may work with older Flutter version but not guaranteed

New

• the plugin now requests both speech and microphone permission on initialize on iOS
• added debugLogging parameter to the initialize method to control native logging

Fix

• The Android implementation now blocks duplicate results notifications. It appears that at least on some Android versions the final results notification onResults is notified twice when Android automatically terminates the session due to a pause time. The de-duplication looks for successive notifications with < 100 ms between them and blocks the second. If you miss any onResult notifications please post an issue.

1.1.0

New

• error_timeout has been separated into error_network_timeout and error_speech_timeout

1.0.0

New

• hasPermission to check for the current permission without bringing up the system dialog
• listen has a new optional cancelOnError parameter to support automatically canceling a listening session on a permanent error.
• listen has a new optional partialResults parameter that controls whether the callback receives partial or only final results.

0.8.0

New

• speech recognizer now exposes multiple possible transcriptions for each recognized speech
• alternates list on SpeechRecognitionResult exposes alternate transcriptions of voice
• confidence on SpeechRecognitionResult gives an estimate of confidence in the transcription
• isConfident on SpeechRecognitionResult supports testing confidence
• hasConfidenceRating on SpeechRecognitionResult indicates if confidence was provided from the device
• new SpeechRecognitionWords class gives details on per transcription words and confidence

Fix

• speechRecognizer availabilityDidChange was crashing if invoked due to an invalid parameter type
• Added iOS platform 10 to example Podfile to resolve compilation warnings

0.7.2

Breaking

• Upgrade Swift to version 5 to match Flutter. Projects using this plugin must now switch to 5.

0.7.1

Fix

• Upgrade Kotlin to 1.3.5 to match the Flutter 1.12 version
• Upgrade Gradle build to 3.5.0 to match the Flutter 1.12 version
• Android version of the plugin was repeating the system default locale in the locales list

0.7.0

New

• locales method returns the list of available languages for speech
• new optional localeId parameter on listen method supports choosing the comprehension language separately from the current system locale.

Breaking

• cancel and stop are now async

0.6.3

Fix

• request permission fix on Android to ensure it doesn't conflict with other requests

0.6.2

Fix

• channel invoke wasn't being done on the main thread in iOS

0.6.1

Fix

• listening sound was failing due to timing, now uses play and record mode on iOS.

0.6.0

Breaking

• The filenames for the optional sounds for iOS have changed.

New

• Added an optional listenFor parameter to set a max duration to listen for speech and then automatically cancel.

Fix

• Was failing to play sounds because of record mode. Now plays sounds before going into record mode and after coming out.
• Status listener was being ignored, now properly notifies on status changes.

0.5.1

• Fixes a problem where the recognizer left the AVAudioSession in record mode which meant that subsequent sounds couldn't be played.

0.5.0

Initial draft with limited functionality, supports:

• initializing speech recognition
• asking the user for permission if required
• listening for recognized speech
• canceling the current recognition session
• stopping the current recognition session
• Android and iOS 10+ support

Missing:

• some error handling
• testing across multiple OS versions
• and more, to be discovered...