Tips for AI Interactive integration with online speech synthesis capabilities

Online speech synthesis is to convert text into sound in a networked scenario, enabling the machine to interact with the human voice. This concept should be a good understanding, the following combined with the official website of the Android online demo to explain the synthesis process and some of the problems you often encounter.

To the official Website SDK Download Center Download the online command word Recognition SDK (for example, the Android version here), you can find that the directory structure inside the package and dictation is very similar, here to tell you a small secret:

Download the SDK package for Android on the official website is identical, why is this? Because dictation, online synthesis, online command word three features in our platform is placed in an SDK package, version is also to do a unified maintenance, the SDK package of the demo is also a collection of this three functions of the demo function (of course, also contains some other features, such as semantic understanding, evaluation), as shown in:

The Java files in the red box are from top to bottom: command word recognition demo, dictation demo, Evaluation sub Demo, demo main portal interface, speechutility initialization, synth demo, semantic interpretation sub demo. It may be noted that the command word recognition demo, dictation sub-demo, synthetic sub-demo all have corresponding local functions, these local features are provided by the language Memory app, not the offline SDK, so if you do not have the language app installed on your phone, The local features you see in the demo are not available (error voice not installed 21001), this part wants to have a further understanding of the post:

21001. Explanation: http://bbs.xfyun.cn/forum.php?mod=viewthread&tid=11724&fromuid=33982

Take a look at the Ttsdemo.java inside the call flow, note that before this must not forget the Speechapp.java inside the speechutility initialization.

* * First is to initialize the composition object * *

MTts = Speechsynthesizer.createsynthesizer (Ttsdemo.this, Mttsinitlistener);

Mttsinitlistener is the initial listener, and it is safer to do the following after the initialization callback succeeds.

* * Then set the relevant composition parameters * *

//Empty Parameters
Mtts.setparameter (speechconstant.params, NULL),
//set corresponding parameters according to the composition Engine
if (menginetype.equals ( Speechconstant.type_cloud) {
Mtts.setparameter (Speechconstant.engine_type, Speechconstant.type_cloud);
// Set online synth pronunciation person
mtts.setparameter (speechconstant.voice_name, voicer);
if (!) Neutral ". Equals (Emot)) {
//current only pronunciation people" AI "Support settings emotion
//" AI "pronunciation person needs to pay for use, please contact: [email protected]
Mtts.setparameter (Speechconstant.emot, Emot);
}
//Set composition speed
Mtts.setparameter (speechconstant.speed, msharedpreferences.getstring ("Speed_preference", " 50 "));
//Set composition Tone
Mtts.setparameter (Speechconstant.pitch, msharedpreferences.getstring ("Pitch_preference", "" "));
Set the composition volume
Mtts.setparameter (Speechconstant.volume, msharedpreferences.getstring ("Volume_preference", ")");

Set the player audio stream type
Mtts.setparameter (Speechconstant.stream_type, msharedpreferences.getstring ("Stream_preference", "3"));
Set Play composition audio interrupt music playback, default to True
Mtts.setparameter (Speechconstant.key_request_focus, "true");
Set audio save path, save audio format support PCM, WAV, set path to SD card please note Write_external_storage permissions
Note: Audio_format parameter requires an updated version to take effect
Mtts.setparameter (Speechconstant.audio_format, "wav");
Mtts.setparameter (Speechconstant.tts_audio_path, environment.getexternalstoragedirectory () + "/msc/tts.wav");

The role of the relevant parameters has been described in more detail in the comments.
* * Final start of synthesis * *
Mtts.startspeaking (text, mttslistener);

Where text is what you need to synthesize, mttslistener is a synthetic listener that contains callbacks that are useful for crafting progress, compositing, and so on.

* * Frequently Asked questions and answers: * *

1, the pronunciation of the parameters can be set to what pronunciation people?

The currently supported online pronunciation list is shown in the following post
"Official" synthetic pronunciation of people online, offline support
http://bbs.xfyun.cn/forum.php?mod=viewthread&tid=12012&fromuid=33982
(Source: Voice Cloud Community)

2, how long can the text be synthesized? What should I do if I want to synthesize text in a very long time?

Synthetic text is a length-limited, one-time maximum of 8,192 bytes of text, and you may be able to convert the maximum number of characters supported according to the encoding format of your text. If the text you want to synthesize is too long, you can divide it by the length limit and then merge or play the audio data in multiple compositions.

3. Is the audio data synthesized on line not compressed? Will it be very expensive to flow?

and dictation when the audio data upload similar, the composition is the server returned audio data is "Speex encoded" compression, compression ratio of about 1:10, that is, when the service side of the synthesized audio 1:10 compression, and then the Client SDK received audio and then decoded into non-compressed raw audio, Then according to the 16K sampling rate of 16-bit sampling precision mono audio to calculate, 1 seconds of audio compression size about 3K, not very large. In addition, the above-mentioned Speex encoding is not a common Speex encoding, so the two ways to compress the audio can not be common.

Tips for AI Interactive integration with online speech synthesis capabilities