audio out not being transfered to the selected nvda audio output

[crucio2211]

hello, can you take a look of this problem?
when I am connecting my computer to a bluetooth speaker and setting nvda to use the speakers on my laptop instead of transfering to bluetooth speaker/earphones, it seems that the eloquence synthesiser doesn't respect my audio output setting on NVDA.
can you see what's the reason behind this?
thank you.

[amirsol81]

Interesting. But I have no issues with Eloquence 64-bit and my Airpods 3 Pro on Windows 11 Pro 25H2.

[fastfinge]

I experience this issue on my Microsoft Surface Laptop, but not my AMD Framework Laptop. I think it has something to do with sound drivers? Either way, restarting NVDA gets the sound device to change.

[crucio2211]

it doesn't work for me.
I am also using windows11 25h2

[harrymkt]

Hello.

I also have the same issue.

• Add-on version: v14.
• Changing the audio device using other TTS engines work and change immediately.
• Restarting does work, but becomes problemmatic if we need to change the device frequently, i.e. in live streams.

[fastfinge]

Is this happening with all addons but eloquence disabled?Sent from my iPhoneOn Mar 1, 2026, at 2:07 AM, Harry Min Khant ***@***.***> wrote:harrymkt left a comment (fastfinge/eloquence_64#58) Hello. I also have the same issue. Add-on version: v14. Changing the audio device using other TTS engines work and change immediately. Restarting does work, but becomes problemmatic if we need to change the device frequently, i.e. in live streams. —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>

[harrymkt]

Yes, only happening on Eloquence. All others are working as expected.

[fastfinge]

I strongly suspect one of your other installed addons, or a sound driver issue. I can only reproduce this on one of four computers I’ve tried. Sent from my iPhoneOn Mar 1, 2026, at 11:44 AM, Harry Min Khant ***@***.***> wrote:harrymkt left a comment (fastfinge/eloquence_64#58) Yes, only happening on Eloquence. All others are working as expected. —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>

[crucio2211]

I try this with eloquence disabled, and the nvda works fine when changing the audio output. On Mon, Mar 2, 2026 at 12:16 AM Samuel Proulx ***@***.***> wrote:

…

*fastfinge* left a comment (fastfinge/eloquence_64#58) <#58 (comment)> Is this happening with all addons but eloquence disabled?Sent from my iPhoneOn Mar 1, 2026, at 2:07 AM, Harry Min Khant ***@***.***> wrote:harrymkt left a comment (fastfinge/eloquence_64#58) Hello. I also have the same issue. Add-on version: v14. Changing the audio device using other TTS engines work and change immediately. Restarting does work, but becomes problemmatic if we need to change the device frequently, i.e. in live streams. —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***> — Reply to this email directly, view it on GitHub <#58 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQKELXUC2M245RXL7ZENA634ORO4DAVCNFSM6AAAAACV4CLATWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTSOBQGM4DENRTGE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[harrymkt]

But Eloquence such as IBM TTS (works in NVDA 2025). And add-ons like Vocalizer, are working properly, meaning they change immediately, even if Eloquence is enabled and not removed.

But when Eloquence is being used as the TTS of the NVDA, the audio device never changes until restart.

I strongly suspect one of your other installed addons, or a sound driver issue. I can only reproduce this on one of four computers I’ve tried. Sent from my iPhoneOn Mar 1, 2026, at 11:44 AM, Harry Min Khant @.***> wrote:harrymkt left a comment (fastfinge/eloquence_64#58)
Yes, only happening on Eloquence. All others are working as expected.

—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: @.***>

[crucio2211]

Yeah true. Tried this with other tts also On Mon, Mar 2, 2026, 8:34 AM Harry Min Khant ***@***.***> wrote:

…

*harrymkt* left a comment (fastfinge/eloquence_64#58) <#58 (comment)> But Eloquence such as IBM TTS (works in NVDA 2025). And add-ons like Vocalizer, are working properly, meaning they change immediately, even if Eloquence is enabled and not removed. But when Eloquence is being used as the TTS of the NVDA, the audio device never changes until restart. I strongly suspect one of your other installed addons, or a sound driver issue. I can only reproduce this on one of four computers I’ve tried. Sent from my iPhoneOn Mar 1, 2026, at 11:44 AM, Harry Min Khant *@*.***> wrote:harrymkt left a comment (fastfinge/eloquence_64#58 <#58>) Yes, only happening on Eloquence. All others are working as expected. —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: *@*.***> — Reply to this email directly, view it on GitHub <#58 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQKELXXUGQZWR2CQ2ONJCG34OTJKDAVCNFSM6AAAAACV4CLATWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTSOBRGQYTMMZWGI> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[fastfinge]

Unfortunately I can't fix this unless I can reliably reproduce it.

[crucio2211]

hi, I have now fix the bug of eloquence when changing the audio output device, atatched here is the files I've edited.
hoping that you will include this in the next build.

The Bug:
When changing the NVDA audio output device (e.g. switching from Bluetooth headphones back to laptop speakers), Eloquence does not follow the new device — it keeps playing audio to the old one. Other synths like OneCore work fine. You mentioned you couldn't reproduce it, so hopefully the explanation below helps.

Root Cause:

The core issue is in SynthDriver.terminate() in eloquence.py — it never calls _eloquence.terminate().

Here's what happens when NVDA reloads the synth driver after an audio device change:

1. NVDA calls SynthDriver.terminate() on the old instance
2. NVDA calls SynthDriver.init() on the new instance
3. init calls _eloquence.initialize() → _client.initialize_audio()
4. initialize_audio() checks if self._player: return — and returns immediately without creating a new player

The problem: _eloquence is a module-level singleton. Python does not reload it when the driver reloads — the same _client object stays in memory with the same _player still pointing at the old audio device. Because terminate() never calls _eloquence.terminate(), the old player is never closed, and initialize_audio() always skips creating a new one.

Files Changed:

synthDrivers/eloquence.py — The main fix (one call):

• Added _eloquence.terminate() at the start of SynthDriver.terminate() so the old WavePlayer is properly closed before the driver reloads.

synthDrivers/_eloquence.py — Supporting fixes:

• Added self._player_lock = threading.RLock() to EloquenceHostClient.init() — it was used in stop() but never defined in the client class, causing an AttributeError crash on every cancel/speech interrupt.
• Fixed EloquenceHostClient.stop() to grab the player reference under the lock before calling .stop(), preventing a race condition crash (OSError: access violation).
• Fixed EloquenceHostClient.shutdown() to wrap self._player.close() in try/except/finally so self._player is always set to None even if .close() throws — previously, an exception here would leave _player non-None, causing initialize_audio() to skip creating a new player on the next load.
• Added a fallback in _create_player() to use the default audio device if the configured device name fails to open.

I've attached the two modified files. Hope this helps for the next build! Let me know if you need any clarification.

eloquence.py

_eloquence.py

[crucio2211]

Hi again! I have a small performance improvement to share, specifically for navigation and typing responsiveness.

The Change:
In synthDrivers/_eloquence.py, three IPC commands were changed from blocking (wait=True) to fire-and-forget (wait=False):

• addText in speak()
• insertIndex in index()
• setVoiceParam in setVParam() when temporary=True (used for caps pitch and prosody)

Why it helps:
Because Eloquence uses a separate 32-bit host process (due to ECI.DLL being 32-bit), every command normally requires a full IPC round-trip — send the command, then block and wait for the host to reply before continuing. For a single speech sequence, this means 3-4 blocking round-trips before synthesis even starts.

However, addText, insertIndex, and temporary setVoiceParam don't actually return any useful data — the host just acknowledges them. So there's no reason to block and wait. By making them fire-and-forget, only the final synthesize command needs to wait for a response, cutting blocking round-trips from ~4 down to 1 per utterance.

In practice, this makes navigation (arrow keys, tab) and typing feel noticeably more responsive.

Risk level:
Low. The host processes commands in order on its end, so fire-and-forget commands will still be executed in the correct sequence before synthesize is handled. The only theoretical risk is if the host's command queue fills up under extreme load, but that's unlikely in normal use.

Please test this and let me know if you spot any issues. Happy to revert if it causes problems!

Attaching the modified _eloquence.py.

_eloquence.py

[fastfinge]

Change made, thanks!