Text to speech in Python | Hacker News Reader

Text to speech in Python

pythonprogramminglanguage.com

113 pointslobdev22 comments

The more accurate title for this is "Python interface to Google's cloud TTS service".

Which means it won't work offline or when Google inevitably shuts down the API.

Yeah, I'm getting a bit of fatigue from all of this "brand new voice assistant!" and "amazing text to speech and speech recognition!" crap which just turns out to be the same wrapper script around Google's APIs that everyone wrote in five minutes when the APIs opened up.

I'm wondering why the Linux TTS offering is so dismal, and why the voices used by Apple, Google and MS are so old. We have seen amazing demos with WaveNet from DeepMind a year ago. Where are the new amazing voices?

Because WaveNet is very impractical in production (per sample prediction). Tacotron may be a better fit (https://google.github.io/tacotron).

Not only that, but you don't get word level timestamps when rendering a phrase. If you want to track speech word by word, it's not possible.

Yeah. Its remember me similar script written for PyS60 for use Symbian 9.x in-built TTS

In Pythonista for iOS:

  import speech

  speech.say('Hola mundo', 'es_ES')

In reverse, to record a sound for three seconds:

  import sound

  r = sound.Recorder('audio.m4a')
  r.record(3)  # seconds

To recognize it as text:

  text = speech.recognize('audio.m4a', 'en')[0][0]  # sent to Apple servers

sudo pip install? Really?

You almost never want to use the system Python. Use pyenv[0] to install the version of Python you want, and then create a new virtual environment for each project.

[0]https://github.com/pyenv/pyenv

That is addressed right before `sudo pip install`.

FTA: "Install using pyenv, pipenv or virtualenv.

If you are feeling brave, install with pip:"

I can't think of a situation where plain "pip install" doesn't work that the correct fix is to use sudo.

I strongly think that if you're going to give advice on how to install, then you'd be better off suggesting:

  python3 -m venv speech-venv
  ./speech-venv/bin/pip install ...
  # or activate and use "pip install"

See also: pipenv

http://docs.pipenv.org/en/latest/

you need sudo for PyCryto or you need to su root b/c when it compiles on linux it needs to write to tmp directories that normal users cannot.

What are your thoughts on conda instead of virtualenv?

Not the person you asked, but I personally don't care for conda.

I'd rather use standard Python packaging and virtual environments. Conda had a few "funnies", like adding lines to .bashrc interfering with system Python when I tried to use it later (easily fixed and not really a big deal, I just didn't care for it).

A client had some scripts that used it, and from that perspective I guess it saved a little time installing everything but not much. Not enough to have another Python version floating around in my opinion. Easy enough to just use regular Python.

So what is the state of the art in turning text into speech files in Python? Especially in terms of free options? Everything I find on google seems to be a year or two out of date.

I'm disappointed - this isn't actually text-to-speech in Python, it's text-to-speech in C++ (or Objective-C in one case) with a Python wrapper over the top.

An actual Python TTS engine would have been cool as an introduction to how the algorithms work without needing to learn C++. The performance would always be lacking but that'd not really matter for an introduction.

I had seen this library a while ago and blogged about it:

PyTTSx: cross-platform Python speech synthesis wrapper

https://jugad2.blogspot.in/2013/01/pyttsx-cross-platform-pyt...

Speech synthesis in Python with pyttsx:

https://jugad2.blogspot.in/2014/03/speech-synthesis-in-pytho...

The synthesized voice announcing a train arrival is cool (link in 2nd post above).

Anyone have an opinion on the best option these days to do the reverse (speech to text)? Building an open source version of Alexa sounds like a fun project.

BadPrototypeError: Objective-C expects 1 arguments, Python argument has 3 arguments for <unbound selector setProperty of NSSpeechDriver at 0x10d9e4e00>

Got that too. But if you're on Mac, you can play around with text to speech using os.

  import os
  os.system("say 'hello world'")

Or anywhere you can install the GNUstep toolkit.

Try virtualenv instead of sudo pip install. Then you can keep track of your requirements.txt, which is helpful for teamwork on multiple machines.

Reply on news.ycombinator.com