mactts is a small Golang wrapper for the Mac OSX Speech Synthesis Manager Carbon API. This was originally created for a TTS speech server demo and as a golang http exercise. While newer versions of Mac OSX ship with the typical Nuance Vocalizer® voices available from many places, there is no matching the utility of the original MacInTalk Pro voices. When you need them and don't have a real Mac handy, what is one to do?

The server command is called gomitalk-server and it is go gettable:

go get github.com/jkl1337/mactts/gomitalk-server

The command gomitalk-server can now be run from the command line. The default port is 8080.

Now go to (http://localhost:8080/say?text=Hello+World&voice=Hysterical) in your browser. You should get a 22,050 Hz 16-bit WAVE file sent to your browser.

The Mac OSX speech synthesis manager does not produce bit identical audio streams from one synthesis to the next. Precisely, it produces the same waveform data but there is an alignment difference of a few samples. This could probably be remedied by using AudioUnits and wrapping the audio to the nearest non-zero sample. Unless this is done, the server cannot be used for serving byte range requests. In order to support this now, a reverse proxy that can cache the output is required. To assist with this the server will optionally (default on) produce Etags that are computed on the selected voice and content type.

The server API supports two endpoints, one for speech and one for information about the voices available:

GET /say{?text,lang,gender,samplerate,type,attachment}
POST /say (application/x-www-form-urlencoded or multipart/form-data)

text : The UTF-8 encoded message text to synthesize as speech.
lang : A locale identifier {language_territory} such as en-US, en-GB that is used to match against the
       available voices. If no match is found, a 404 is returned.
       Currently, only the language-territory format is allowed.
gender : A gender name {male,female,neuter} used to match against available voices. If this is
         specified and no match is found, a 404 is returned.
samplerate : One of (8000, 11025, 16000, 32000, 44100, 48000). If no match is found, 22050 is used.
type : The preferred MIME type of the audio. Either audio/wav or audio/mp4. Some equivalent variants of
       these are allowed. This is provided because most browsers such as Chrome and Firefox do not use the
       type attribute of the <audio> element to set the Accept header in such a way that the preferred
       audio type is retrieved.
attachment: A filename that is used to set the Content-Disposition header.


GET /voices

Returns a JSON object with names, languages, and genders of available voices on the system.

• Provide MP4 and Vorbis output so the audio tag can be used on all browsers with compression.