The bounded lock wait was 10 x 0.2s = 2s, but the concurrent-discovery
scenario this lock exists for (#62771: hermes serve + gateway spawning
every MCP server twice) reports 40-60s discovery rounds. A 2s budget
guarantees the loser times out and runs unguarded exactly when the guard
matters. Widen to 240 x 0.5s = 120s; fail-soft unguarded fallback and
test overrides unchanged.
Detached background delegation batches (_batch_runner) no longer honor
the foreground parent's interrupt flag — a busy-submit interrupt in the
TUI/desktop previously fabricated 'interrupted' results for background
children that should outlive the turn. Explicit cancellation still works
via _batch_interrupt.
Rebased onto current main from PR #65040; both interrupt-suppression
regression tests aligned with the current _session test helper.
Original work by @AtakanGs in #65040.
llama_cpp's GGUF backbone loader only enables n_gpu_layers for the
literal device string "gpu"; torch's codec only accepts "cuda". A
single --device value passed straight through can't satisfy both,
so config.yaml's documented tts.neutts.device: cuda silently ran the
backbone on CPU while the codec ran on GPU.
Map cuda -> gpu for the backbone only; leave the codec on the
torch-native string. Measured 24.9s -> 13.1s per synthesis call on
an RTX 5070 Ti (neutts-air-q4-gguf) with this fix.
`_play_via_tempfile` passes the NamedTemporaryFile *object* to `wave.open()`.
`wave.open()` flushes but does not close a file it did not open itself (by
name), so the OS handle to the temp WAV stays open. On Windows that open write
handle blocks the system player from reading the file and blocks the
`os.unlink()` cleanup (WinError 32, silently swallowed by `except OSError:
pass`), leaving orphaned temp .wav files piling up. The fallback runs whenever
no sounddevice output stream is available (sounddevice not installed / no audio
device), which is common on headless and Windows setups.
Close the temp file handle before invoking the player and again in the finally
block (idempotent), so the player can read the file and `os.unlink()` can
remove it.
Regression test drives `stream_tts_to_speaker` with the sounddevice path forced
unavailable and asserts the temp handle is closed before `play_audio_file` is
called (mutation-verified: reverting the close leaves the handle open at play
time and the test fails). Cross-platform.
Concurrent TTS requests within the same second (e.g. from voice prefetch)
generated identical tts_YYYYMMDD_HHMMSS.ogg paths, causing a race where
one request would unlink the file while another was still writing — resulting
in 'produced no output' errors.
Adding %f (microseconds) to the strftime format makes collisions practically
impossible. Registered as PATCH-006.
Cloud STT APIs (Groq, OpenAI, etc.) cannot parse Apple CAF containers.
Add .caf to SUPPORTED_FORMATS and _convert_caf_to_wav() which tries
ffmpeg (cross-platform) then afconvert (macOS built-in).
Two in-house micro-fixes from issue triage:
- #49883: voice.beep_enabled gates in cli.py and hermes_cli/voice.py used
bool() on the config value, so a quoted YAML string like "false" or
"off" kept beeps on. Route through utils.is_truthy_value.
- #18432: AudioRecorder.start() collapsed OSError from _import_audio into
the 'pip install sounddevice numpy' hint — but OSError means the
PortAudio SHARED LIBRARY is missing, which pip cannot fix. Mirror
detect_audio_environment's system-package hint (libportaudio2 /
brew portaudio / Termux pkg install portaudio) on that path.
Fixes#49883Fixes#18432
When Hermes runs inside its bundled venv (~/.hermes/profiles/<name>/
hermes-agent/venv/), the bare 'pip install sounddevice numpy' hint in
_voice_capture_install_hint() told users to install into whichever
Python their shell resolves first — on macOS that is often the system
Python under Rosetta, a completely separate site-packages tree with
incompatible wheel arches.
Detect venv via sys.prefix != sys.base_prefix and return the full path
to the venv's pip binary (Path(sys.prefix)/bin/pip) when available.
Falls back to the bare hint outside a venv (e.g. bare-metal installs).
Also adds 'from pathlib import Path' which was the only new import.
Addresses review on #62601. Applies a single rule — no sounddevice for
audio OUTPUT on macOS (PortAudio/CoreAudio init triggers a
kTCCServiceMediaLibrary prompt) — consistently at all three output sites:
- play_audio_file: WAV playback (already routed to afplay) now uses the
shared _sounddevice_output_allowed() helper.
- play_beep: synthesize the tone with numpy only, then on macOS play it
via a temp WAV through afplay instead of sounddevice.
- stream_tts_to_speaker (tts_tool): on macOS, skip the sounddevice
OutputStream so playback falls through to the existing tempfile/afplay path.
Audio INPUT (recording) is untouched — it legitimately needs mic permission.
Tests: TestMacOSAudioOutputPolicy (voice_mode) proves WAV + beep routing
does not import sounddevice on Darwin and still uses it off Darwin;
test_tts_macos_output proves streaming TTS skips the OutputStream on Darwin.
Existing test_play_wav_via_sounddevice pinned to non-Darwin for determinism.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
On macOS, initializing PortAudio/CoreAudio via sounddevice triggers a
kTCCServiceMediaLibrary permission dialog even when no media-library
access is needed. afplay already handles WAV (and every other format)
natively, so route macOS playback straight to it and keep sounddevice
for other platforms.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
\`pm list packages com.termux.api\` is the canonical way to detect the
Termux:API Android app, but on some devices it gives a false negative
even when the app is installed and \`termux-microphone-record\` runs
fine — the symptom reported in #31015 (\`/voice on\` complaining
"Termux:API Android app is not installed").
Replace the single probe with a graded strategy:
1. Try \`pm list packages com.termux.api\` (current behaviour).
2. If \`pm\` isn't on PATH or returns non-zero, fall back to
\`cmd package list packages com.termux.api\` — the modern Android
API 28+ equivalent that's present on devices where \`pm\` is gone.
3. If both probes are inconclusive (binary missing, permission
denied, timeout, or non-zero exit) and \`termux-microphone-record\`
is on PATH, trust the binary. The CLI ships in the \`termux-api\`
package which is essentially only useful with the Android app
installed; users who installed it deliberately almost always have
the app too.
Polarity matters: a false negative on this gate blocks \`/voice on\`
entirely (the user-reported symptom), while a false positive only
surfaces a precise runtime error from the binary itself when it
tries to talk to the missing app — strictly more actionable.
The clean-probe-but-no-package case still returns False, so the
existing "Termux:API CLI installed without the app" warning still
fires when the package manager *can* tell us the app is missing.
Refs: NousResearch/hermes-agent#31015
Ports #63768 forward onto current main per teknium1's review.
On WSL2 without a PulseAudio bridge, ffplay and aplay have no audio
device and TTS playback silently fails (issue #17608). When
powershell.exe and ffmpeg are available, convert the audio to a
uniquely-named WAV in the Windows %TEMP% directory and play it via
Media.SoundPlayer.
Per review, this fixes two gaps in the original port:
1. Exit-status masking: the cleanup subshell was
'( ffmpeg && powershell ); rm -f wav' -- the shell's exit status is
the LAST command's (rm -f, which is always 0), so a real
ffmpeg/PowerShell failure could never be detected by the rc-checking
fallback logic added to the player loop. Now captures the real
status before cleanup and re-exits with it:
'( ffmpeg && powershell ); rc=0; rm -f wav; exit '.
2. The no-Pulse WSL gate in detect_audio_environment() still hard-blocked
voice mode entirely (input AND output) even when the PowerShell
fallback made TTS output viable. Added _wsl_powershell_tts_available()
and use it to downgrade the WSL-without-Pulse case from a hard
'warnings' block to a non-blocking 'notices' entry when the fallback
is available -- the same PulseAudio-bridge recording guidance is still
surfaced (mic capture genuinely still needs it), it just no longer
blocks /voice on for TTS-only usage. cli.py's existing
env_check['available'] gate needed no changes since it already
respects this flag.
Also fixed the flaky uniqueness test (the original asserted
len(filenames) >= 2, which passed trivially on zero captured
filenames) and added a real fallback-triggering regression test for
the exit-status fix.
10 new/fixed tests pass in TestWSL2PowerShellFallback and the new
TestWSLAudioEnvironmentGate; 80/80 in the full tests/tools/test_voice_mode.py file.
WSLg RDP audio has two issues causing crackling:
1. systemd-timesyncd clock adjustments jitter PulseAudio timing
(microsoft/wslg#1257) — user action: stop the service
2. Cold-start RDP connection drops first ~100ms of audio packets
before the virtual channel stabilises
Fix (automated, WSL-only):
- Detect WSL via /proc/version 'microsoft' marker
- Prepend 100ms silence + apply 100ms fade-in to audio
- Append 50ms silence tail for clean stream teardown
- Set blocksize=4096 (default auto ~1024 is too small for RDP)
- All in a single continuous sd.play() buffer
Non-WSL paths unchanged.
Closes#38893
AudioRecorder hard-codes SAMPLE_RATE (16 kHz) when opening the input
stream, but some capture devices (e.g. USB microphones exposed through
ALSA hw) reject 16 kHz outright — sd.InputStream fails with
PaErrorCode -9997 (Invalid sample rate) and voice recording is broken.
Query the default input device for its native default_samplerate at
recording start and open the stream / write the WAV at that rate,
falling back to the Whisper-friendly 16 kHz constant when the backend
does not expose a usable rate. STT providers accept standard WAV rates,
so downstream transcription is unaffected.
Closes#55908. The CLI voice-mode beep amplitude is hardcoded at 0.3 inside
tools.voice_mode:play_beep(), which makes the record start/stop cues too
quiet on low-volume systems and headphones. Users couldn't adjust it
without editing source.
Move the literal into a configurable voice.beep_volume setting (clamped to
0.0-1.0, default 0.3 to preserve prior behaviour). The new
_get_beep_volume() helper reads via the same load_config() pattern used by
cli.py's _voice_beeps_enabled() and hermes_cli/voice.py's _beeps_enabled(),
keeps bools / out-of-range / non-numeric / NaN values safely on the default,
and falls back silently if config can't load so the audio cue never breaks
the voice loop on a degenerate config.yaml.
Covered by tests/tools/test_voice_mode.py:
- TestGetBeepVolume (12 cases: missing key, custom value, boundary 0.0/1.0,
out-of-range clamp, type coercion, bool guard, NaN guard, exception
guard, dict-typed voice section)
- TestPlayBeepVolumeWiring (guards against re-introducing a hardcoded 0.3
literal in play_beep)
Docs: website/docs/user-guide/configuration.md mentions the new key.
Other locale translations (zh-Hans etc.) intentionally untouched —
handled by the regular i18n sync pipeline as a separate change.
No change in default behaviour: existing users hear exactly the same beep.
Salvaged from PR #18861 (@shellybotmoyer). Switch the central Opus
transcoders from CBR 64k (-vbr off) to VBR 48k with -application voip and
-compression_level 10. The old CBR settings produced lower-quality speech
audio and occasionally failed to trigger Telegram's native voice-bubble
rendering. Applied to _ffmpeg_transcode_to_opus (the single transcoder
behind _convert_to_opus and _repair_ogg_container), the Gemini WAV→OGG
path, and the Matrix adapter boundary transcoder added in this branch.
Fixes#18818
Consolidates all TTS text-preparation paths onto
tools/tts_text_normalize.prepare_spoken_text:
- strip_nonspoken_blocks: removes <think> reasoning blocks (#34213,
incl. unterminated streaming blocks) and the end-of-turn
file-mutation verifier footer emitted by run_agent.py (#40772).
- flatten_newlines_for_payload: collapses newlines into sentence
breaks so newline-sensitive OpenAI-compatible providers (Kokoro)
speak the whole script instead of truncating at the first newline
(#9004).
- tools/tts_tool._strip_markdown_for_tts (voice-mode streaming + web
dashboard path) now delegates to the shared cleaner, with the legacy
regex pipeline kept as a best-effort fallback.
- hermes_cli/voice.py speak_text and cli.py _voice_speak_response now
use the shared cleaner instead of their own duplicated regex
pipelines.
- gateway auto-TTS fallback also strips think blocks.
Tests: tests/tools/test_tts_prepare_spoken.py covers think blocks,
verifier footer, emoji, newline flattening, and the shared-cleaner
wiring on the tool/streaming/gateway paths. Updated the header
expectation in test_voice_cli_integration.py for the heading-fold
behavior of the shared cleaner.
Closes#34213, #9004, #40772
The 'language' config field in tts.openai.language was read but never passed
to the API. This caused Kokoro (and other OpenAI-compatible TTS backends that
support lang_code) to default to English phonemization regardless of the
configured language.
Now passes lang_code via extra_body when language is set in config.
xAI TTS supports a text_normalization boolean that normalizes
written-form text (numbers, abbreviations, symbols) into spoken-form
before synthesis. This parameter was not being sent, leaving users
with literal number/symbol pronunciation.
- Add DEFAULT_XAI_TEXT_NORMALIZATION_DEFAULT = False
- Read tts.xai.text_normalization from config (via _xai_bool_config)
- Attach text_normalization: true to the POST payload when enabled
- Omit the field entirely when unset or explicitly false (API default)
- Add 3 tests: default omission, enabled passthrough, explicit false
Adds an optional provider parameter to the text_to_speech tool that lets the model select a TTS provider per-call instead of always using the globally configured tts.provider.
When provider is set, it bypasses the configured default and routes directly to the specified backend.
When omitted (the default), the tool behaves exactly as before.
Closes#47459
The `text_to_speech` tool schema accepted only `text` and `output_path`,
so style direction (tone, emotion, pacing, whispering) could never reach
the OpenAI backend — even though `gpt-4o-mini-tts` (Hermes's OpenAI
provider default) treats `instructions` as its primary voice-design
control.
This plumbs an optional `instructions` argument through the tool schema,
the handler lambda, and `text_to_speech_tool()` into
`_generate_openai_tts`, where it is forwarded to
`client.audio.speech.create()` only when truthy. Empty/None values still
omit the key entirely, preserving behavior on `tts-1`/`tts-1-hd` and
strict OpenAI-compatible servers.
The same passthrough unblocks self-hosted OpenAI-compatible voice-design
servers (Qwen3-TTS-VoiceDesign on oMLX, etc.) that are already wired in
via `tts.openai.base_url` — the established convention per #9004 and the
TTS config docs — without inventing a new provider backend.
Tests: `tests/tools/test_tts_instructions.py` covers backend passthrough,
tool-level threading, schema declaration, and the empty-string/absent
omission cases. `tests/tools/test_tts_max_text_length.py` fake_openai
signature widened to accept the new kwarg.
Refs NousResearch/hermes-agent#14196
Add optional 'speed' parameter (0.25-4.0) to the text_to_speech tool
schema and handler. When provided by the model, it overrides the
config-level tts.speed setting, enabling per-request speed control
without config changes.
Use cases:
- Language learning: slow playback (0.5x) for pronunciation practice
- Accessibility: adjustable speed for hearing preferences
- Content review: accelerated playback for long text
The speed value is clamped to [0.25, 4.0] and injected into tts_config
before dispatching to any provider (Edge, OpenAI, MiniMax, etc.), so
all existing provider-level speed handling works transparently.
Includes 3 new tests for tool-level speed injection, clamping, and
config preservation when speed is not specified.
When /reasoning show is enabled, the model's <think> blocks appear in
the final assistant message. TTS reads these aloud, which is unwanted —
users want to see reasoning but not hear it spoken.
- Add _THINK_BLOCK regex to _strip_markdown_for_tts() (tools/tts_tool.py)
— the general TTS text-preparation path used by all TTS providers
- Add <think> stripping to prepare_tts_text() (gateway/platforms/base.py)
— the gateway auto-TTS path
- Reuses the existing regex pattern from stream_tts_to_speaker()
Closes#34213
Auto-TTS previously fed raw chat Markdown and compact symbols straight to the
speech provider, so units were read as stray letters and headings or bullets ran
together. This routes spoken text through a new normalizer,
tools/tts_text_normalize.prepare_spoken_text, that expands units (for example a
temperature written with the degree symbol becomes "degrees Celsius") and
flattens Markdown into a transcript-like script with sentence pauses.
The normalizer is best-effort: if it ever fails the code falls back to the
previous markdown-strip behavior, so auto-TTS keeps working. The Telegram voice
caption uses the same normalized text. Includes a unit test.
Salvaged from PR #62977 (@Vissirexa) — TTS model-cache half only (the
hindsight turn-buffer half is a different subsystem and was dropped).
_piper_voice_cache and _kittentts_model_cache were keyed by voice/model
with no eviction, and each entry is a whole loaded model (tens of MB).
A surface that sweeps voices pinned one model per voice for the process
lifetime. New _tts_cache_get_or_load() get-or-loads through a small LRU
(_TTS_MODEL_CACHE_MAX=3), refreshing recency on a hit and evicting the
least-recently-used model on a cold miss.
Class-level sweep following the ElevenLabs salvage (#66311): every cloud
TTS provider section now honors tts.<provider>.base_url. xAI, MiniMax,
Gemini, OpenAI and DeepInfra already did; Mistral (SDK server_url) was
the remaining gap. Adds per-provider config tests locking in the
ElevenLabs environment plumbing and the Mistral server_url passthrough.
Salvaged from PR #66311 (@moeadham), rebased onto the current streaming
registry. tts.elevenlabs.base_url (+ optional wss_url, derived from
base_url when omitted) routes both the sync ElevenLabs path and the
chunked ElevenLabsStreamer through an ElevenLabsEnvironment, matching
the STT side's ELEVENLABS_STT_BASE_URL/config override pattern.
Follow-up to salvaged PR #70307 (@aml1973): OpenAIStreamer now checks
tts.openai.api_key (config.yaml) ahead of the env resolver in both
available() and stream(), completing config parity between the sync
and streaming OpenAI TTS paths.
Use the shared OpenAI audio key resolver and prefer tts.openai.base_url over the global environment fallback in the Desktop streaming path. Add focused regression coverage for credential and endpoint propagation.
Salvaged from PR #26233 (@LeonSGP43), rebased onto the current 3-tuple
_resolve_openai_audio_client_config (is_managed flag, post-#73072 layout).
Same fix independently submitted earlier in PR #26209 (@zccyman) — credit
to both.
Resolution order now mirrors the STT resolver: tts.openai.api_key/base_url
from config.yaml -> VOICE_TOOLS_OPENAI_KEY/OPENAI_API_KEY env (still
honoring config base_url) -> managed gateway. _has_openai_audio_backend
also counts a config api_key as an available backend.
Fixes#26175
- parametrize the OAuth retry regression over HTTP 401 and 403 so both
documented rejection statuses are exercised
- reword the retry-failure warning: the except block also covers the
retried request, not just the credential refresh
- drop docs/proof/ from the PR diff; the live-proof screenshot now lives
on the fork's proof-assets-xai-stt-oauth-retry branch and stays linked
from the PR description
Follow-ups for the salvaged wave: the auto-detect legacy-error test now
stubs the split validators, the unknown-command-provider test expects
the new provider_not_registered error, and _transcribe_local tolerates
a null stt.local config section again.
Two small fresh fixes on top of the salvage wave:
- Wrap the check-then-load of the module-global faster-whisper model in a
double-checked threading.Lock so concurrent voice messages can't both
download/load the model (#24767).
- Treat an empty stt.openai.api_key as no-auth when stt.openai.base_url
points at a loopback/RFC-1918/.local host, so local OpenAI-compatible
STT servers (faster-whisper-server, speaches, vLLM whisper) work
without a sham api_key value. Reimplements the idea from PR #25193 —
credit @nnnet.
Co-authored-by: nnnet <nnnet@users.noreply.github.com>
Newer OpenAI transcription models (gpt-4o-transcribe, gpt-4o-mini-transcribe)
reject some containers the legacy whisper-1 endpoint accepted -- notably the
Ogg/Opus voice notes messaging platforms deliver -- returning a 400
'corrupted or unsupported' error, so voice-note transcription fails for users
on those models even though SUPPORTED_FORMATS still advertises .ogg/.aac/.flac.
Wrap the OpenAI upload: on a format-related BadRequestError, transcode the
source to a compact 16 kHz mono AAC .m4a via ffmpeg and retry once. This is
model-agnostic (no per-model format table to maintain) and adds no cost for
formats the endpoint already accepts.
Fixes#68719
