Synthesized audio not correct when synthesized within an installed VS code extension

[deinhofer]

I have added some test infrastructure in the extension project, so that I can easily compare the audio output for different providers for the same text, see Running Commandline Tests.

I have also added some debugging messages to the js-tts-wrapper to check, what is actually sent to the API, see PR#27.

If I now compare the log messages for the examples/Simple-Speechmarkdown-Examples.smd, I get two different results:

Run node script locally

Running the node script with the locally linked js-tts-wrapper with node test/src/test-simple-smd.js , I get:

Azure SSML warnings: [
  "Engine 'azure' requires xmlns attribute in <speak> tag.",
  "Engine 'azure' requires version attribute in <speak> tag."
]
AzureTTSClient.synthToBytes - TTS text <speak xml:lang="en-US" version="1.0" xmlns="http://www.w3.org/2001/10/synthesis"><voice name="zh-CN-XiaoxiaoMultilingualNeural"><prosody rate="medium" pitch="medium" volume="100%">MS Azure test
There is a short pause <break time="500ms"/>, before I continue.
I can make text <emphasis level="strong">important</emphasis> or use very emphasised and slightly emphasised.
<prosody rate="x-slow">I can speak text slow</prosody> and <prosody rate="x-fast">I can speak text fast</prosody>.
<prosody pitch="high">I can speak text high</prosody> and <prosody pitch="low">I can speak text low</prosody>.
<prosody volume="loud">I can speak text loud</prosody> and <prosody volume="soft">I can speak text soft</prosody>.</prosody></voice></speak>, Options: {"useSpeechMarkdown":true,"format":"mp3"}

Run within installed extension

When I run it within the installed extension v0.0.20-debug1, I get the following log message:

console.ts:139 [Extension Host] Converting Speech Markdown to SSML for Azure.
console.ts:139 [Extension Host] oA.synthToBytes - TTS text <speak xml:lang="en-US" version="1.0" xmlns="http://www.w3.org/2001/10/synthesis"><voice name="zh-CN-XiaoxiaoMultilingualNeural"><prosody rate="medium" pitch="medium" volume="100%">There is a short pause <break time="500ms"/>, before I continue.
I can make text ++important++ or use (very emphasised)[emphasis:'strong'] and (slightly emphasised)[emphasis:'reduced'].
(I can speak text slow)[rate:'x-slow'] and (I can speak text fast)[rate:'x-fast'].
(I can speak text high)[pitch:"high"] and (I can speak text low)[pitch:"low"].
(I can speak text loud)[volume:"loud"] and (I can speak text soft)[volume:"soft"].</prosody></voice></speak>, Options: {"useSpeechMarkdown":true,"format":"mp3"}

The vs code extension also has speechmarkdown-js as dependency, maybe that is the reason? Or something in the packaging process?
It seems that a behaviour of the peer dependency is different or that you use some dynamic loading of libs that behaves differently, when executed within the extension.

[willwade]

Thanks for the detailed issue. Thats super awesome. I’ve added a regression test for Speech Markdown auto‑detection and a Node createRequire fallback so speechmarkdown-js can load in bundled/extension environments, plus restored auto‑enable in synthToBytesWithConversion/speakStreamed. This should address the extension behavior where the fallback converter was being used. Please retest with the next build. Well I hope so! Give it a whirl. v0.1.65

[deinhofer]

Thank you for the quick update.
There seems to be a syntax error in the new code.
After installing 0.1.65 and trying to run the test-script, I get the following error:

/media/mad/Daten/dev/TTS/speechmarkdown-vscode/node_modules/js-tts-wrapper/cjs/index.js:1373
                    metaUrl = new Function("return "file://" + __filename")();
                                           ^^^^^^^^^

SyntaxError: missing ) after argument list
    at wrapSafe (node:internal/modules/cjs/loader:1662:18)
    at Module._compile (node:internal/modules/cjs/loader:1704:20)
    at Object..js (node:internal/modules/cjs/loader:1895:10)
    at Module.load (node:internal/modules/cjs/loader:1465:32)
    at Function._load (node:internal/modules/cjs/loader:1282:12)
    at TracingChannel.traceSync (node:diagnostics_channel:322:14)
    at wrapModuleLoad (node:internal/modules/cjs/loader:235:24)
    at Module.require (node:internal/modules/cjs/loader:1487:12)
    at require (node:internal/modules/helpers:135:16)
    at Object.<anonymous> (/media/mad/Daten/dev/TTS/speechmarkdown-vscode/test/load-env.cjs:4:29)

Node.js v22.16.0

It obviously need some escaping of the double quotes....
Thank you!

[willwade]

Its half because Im not able to test this fully.. I really really must do better.. Anyway I've just pushed 0.1.66 and I've done some further things in this like updating the dependencies and linting.. if you spot anything do shout. Fingers crossed this fixes your problem. Just open it if doesnt

[deinhofer]

Hi,

Unfortunately, it still does not work.
The log message gives a hint, that speechmarkdown-js was not available during loading:

[Extension Host] Saving TTS output to: /media/mad/Daten/dev/TTS/speechmarkdown-vscode/examples/tts-output/azure_Simple-Speechmarkdown-Examples_2026-01-19_18-46-25.mp3, synthOptions: {useSpeechMarkdown: true}
console.ts:139 [Extension Host] Converting Speech Markdown to SSML for Azure.
console.ts:139 [Extension Host] speechmarkdown-js not available. Using built-in fallback. To enable full Speech Markdown in browsers, add 'speechmarkdown-js' to your app and it will be loaded at runtime.
console.ts:139 [Extension Host] Azure SSML warnings: (2) ["Engine 'azure' requires xmlns attribute in <speak> tag.", "Engine 'azure' requires version attribute in <speak> tag."]
console.ts:139 [Extension Host] oA.synthToBytes - TTS text <speak xml:lang="en-US" version="1.0" xmlns="http://www.w3.org/2001/10/synthesis"><voice name="zh-CN-XiaoxiaoMultilingualNeural"><prosody rate="medium" pitch="medium" volume="100%">There is a short pause <break time="500ms"/>, before I continue.
I can make text ++important++ or use (very emphasised)[emphasis:'strong'] and (slightly emphasised)[emphasis:'reduced'].
(I can speak text slow)[rate:'x-slow'] and (I can speak text fast)[rate:'x-fast'].
(I can speak text high)[pitch:"high"] and (I can speak text low)[pitch:"low"].
(I can speak text loud)[volume:"loud"] and (I can speak text soft)[volume:"soft"].</prosody></voice></speak>, Options: {"useSpeechMarkdown":true,"format":"mp3"}
console.ts:139 [Extension Host] Audio saved to temporary file: /tmp/tts-audio-1768844788490.mp3
console.ts:139 [Extension Host] Playing audio with mpg123...
console.ts:139 [Extension Host] Audio playback process exited with code 0

I cant't reopen the issue because I don't have the permission for it.

[willwade]

Ah I've been hunting the wrong thing. Ok. I give in. Give me a while to get this vs speechmarkdown code running.

[deinhofer]

There is no released npm package 0.1.67 yet.
It seems that the automated build failed: https://github.com/willwade/js-tts-wrapper/actions/runs/21154247165/job/60836040681

[willwade]

@deinhofer thanks for pointing that out! Quick question.. How would you feel if I dropped node 18.x?

[deinhofer]

@deinhofer thanks for pointing that out! Quick question.. How would you feel if I dropped node 18.x?

I am using v22, so no problem, but I don'T know if 18.x is still very common.

[deinhofer]

@willwade how is it going? any progress in this regard? Do know what the actual problem is and why with the extension environment the speechmarkdown support is not enabled properly?

[willwade]

I’ve now (FINALLY!) re-tested this end-to-end with speechmarkdown-vscode and js-tts-wrapper (including latest 0.1.69) using your same SpeechMarkdown sample. In the current builds, the Azure path is producing converted SSML (for example / ), not raw markdown markers, before the API call is made.

I also added a permanent regression test in js-tts-wrapper to catch this exact class of failure (silent fallback instead of full parser conversion), so this should not regress unnoticed again.

Could you please re-test in your extension setup with js-tts-wrapper@0.1.69 and share the current AzureTTSClient.synthToBytes - TTS text ... log line?
If that line still shows raw markers on your side, I’ll use your exact extension packaging/runtime path as the next repro target and fix it directly (well, ill try!)