PoshMouth A few weeks ago I was scanning through my daily PowerTips from PowerShell.com and came across one that covered Using COM Objects to Say Hi. In this tip, they describe how to use the Microsoft Speech API (SAPI) to convert text to play through the the windows audio system.  Of course I started tinkering around to see how SAPI sounded with various text strings.

Once the novelty of that wore off, the next thing I thought of was how I could extend my PowerShell Twitter library PoshTweet with the SAPI library.  I quickly built a little script that called the PoshTweet Get-TwitterFriendsTimeline and passed the status values to this call and it just worked!

While that was a great proof of concept, it's not very practical.  I quickly wrote up the following set of requirements of how I would "like" this little project to come out.

  1. It should work as a centralized service.
  2. The audio should be persistent.
  3. It should be a controlled system.  Running off the public timeline would likely kill my system.

So I bring to you: PoshMouth!

Centralized Service

True to my PowerShell roots, I opted to build the service in PowerShell.  The main body of the script is as follows:

Set-TwitterCredentials -user $TwitterUser -pass $TwitterPass

# Start Application Loop
    # Query the current users rate limit status
    [xml]$rls = Get-TwitterRateLimitStatus -raw $true;
    $remaining_hits = $rls.hash.{remaining-hits}.get_InnerText();
    [DateTime]$reset_time = $rls.hash.{reset-time}.get_InnerText();

    Write-DebugMessage "------------------------------------------------------";
    Write-DebugMessage "Remaining Hits: $remaining_hits...";
    Write-DebugMessage "Reset Time    : $reset_time";
    Write-DebugMessage "------------------------------------------------------";
    if ( $remaining_hits -gt 0 )
        # First pull up all friends timelines
        [xml]$x = Get-TwitterFriendsTimeline -raw $true -since_id (Get-LastStatusId);
        Process-StatusEntries $x;

        # Sleep until next poll
        Write-DebugMessage "Sleeping for $DELAY seconds...";
        Start-Sleep -Seconds $DELAY;
        # Determine how many seconds needed
        $ts = $reset_time - [DateTime]::Now;
        $wait_time = [int]$ts.TotalSeconds + 1;
        # Sleep until new rate limit is allocated
        Write-DebugMessage "Waiting $wait_time seconds for new Rate Limit...";
        Start-Sleep -Seconds $wait_time;

Basically the API Rate Limit is queried to determine if calls can be made.  If so, then Get-TwitterFriendsTimeline function is called to request the current timeline.  Then each entry is processed with the following code:

function Process-StatusEntries()
  param($status = $null);
  if ( $status )
    $status.statuses.status | ForEach-Object -Process {
     # Ignore tweets from this account.
    if ( ! $_.user.screen_name.ToLower().Equals($TwitterUser) )
       # If audio already exists, we'll assume it's already been processed.
       if ( ! (Audio-Exists $_.id) )
         Write-DebugMessage "Processing Tweet # $($_.id)...";   
        $mp3file = Create-AudioFile $_;
         $url = Publish-AudioFile $mp3file;
         Tweet-Notification -status $_ -mp3file $url;
         Set-LastStatusId -id $_.id;

I had to check the screen_name to make sure it didn't match the twitter account this was running under or an infinite loop could occur where I convert PoshMouth's tweets of tweets of tweets...

If the audio already exists in the local repository, then the tweet is skipped.  Otherwise, a MP3 is created and published to a web server.  Finally an @reply tweet is made to the original posters username with a link to the target mp3 file.

Persistent Audio

As you noticed above, to make the audio persistent, I needed to save the audio stream to a file on the system.  The Speech API allows for saving to an uncompressed .WAV file via the SAPI.SpFileStream object.  Since .WAV's are so passe, I figured I'd throw a little LAME action in there and convert that .WAV to a more appropriate .MP3 file.  The following code does the dirty work.

function Convert-TextToWAV()
    param([string]$text = $null, [string]$wavfile = $null);
    if ( $text -and $wavfile )
        $SSMCreateForWrite = 3;
        $DoEvents = 0;
        $SVSFlagsAsync = 1;
        # remove file if it exists
        #Remove-Item -ErrorAction SilentlyContinue $wavfile;
        $spFs = New-Object -ComObject SAPI.SpFileStream
        $spFs.Open($wavfile, $SSMCreateForWrite, $DoEvents)
        $voice = New-Object -ComObject SAPI.SpVoice;
        $voice.AudioOutputStream = $spFs;
        $voice.Speak($text, $SVFlagsAsync);
function Convert-WAVToMP3()
    param([string]$wavfile = $null, [string]$mp3file = $null);
    if ( $wavfile -and $mp3file )
        & $LAME_EXE -S $wavfile $mp3file;
function Convert-TextToMP3()
    param([string]$text = $null, [string]$mp3file = $null);
    if ( $text -and $mp3file )
        $wavfile = Get-TempFile "foo.wav";
        Convert-TextToWAV $text $wavfile;
        Convert-WAVToMP3 $wavfile $mp3file;
        Remove-Item $wavfile;
function Create-AudioFile()
    param($status = $null);
    if ( $status )
        $created_at = $status.created_at;
        $id = $status.id;
        $text = $status.text;
        $user_id = $status.user.id;
        $user_name = $status.user.name;
        $screen_name = $status.user.screen_name;
        $audio_file = Get-AudioFile $id;
        [void](Convert-TextToMP3 $text $audio_file);
    return $audio_file;

The Create-AudioFile function will take the status output from Twitter, and Convert the text to a .MP3 on the local system with the SAPI SpFileStream object along with the publicly available LAME audio library.

Controlled System

The process for converting a tweet to MP3 and uploading to our media server takes about 3-5 seconds.  If I allowed everyone that followed the Twitter account, this could grow way beyond the ability for my little script to handle.  I'd need to build some sort of asynchronous setup to allow for parallel encoding and uploading.  This, in addition to the fact that within 5 minutes of creating the account, I already had 3 followers trying to sell me on weight-loss products, I knew this wasn't an option.  To limit this, I decided to set it up to only create audio files for accounts that my account was following.  This way I could easily control for whom the audio is getting created.

Putting It All Together

So how can you start getting your tweets automagically audioized?  Here's the process:

  1. Start following the @TweetMouth account on Twitter.
  2. Sit back and wait for me to follow you.
  3. Watch for @replies from @TweetMouth.  Click on the tinyurl link and you can hear your tweet in all it's Windows Sound System glory!

Getting the Bits

PoshMouthBot.zip - A zip containing the latest PoshMouthBot.ps1 and PoshTweet.ps1 scripts.

If you decide to build and run your own version of PoshMouthBot, you'll need to use your own twitter account and have access to a distribution server for the MP3 files.  Most of this is controlled from global variables at the top of the PoshMouthBot.ps1 script. 

Enjoy and happy Tweeting!