piperust/.env.example

# Audio Agent Engine — Environment Variables
# Copy to `.env` and adjust as needed.

# --- Server ---
SERVER_PORT=8787

# --- TTS Engine Selection ---
# Choose the TTS engine plugin. Supported values:
#   fish      — Fish Audio / SGLang Drama 3 (OpenAI-compatible HTTP endpoint)
#   kokoro    — Kokoro TTS (stub; see src/tts/kokoro.rs)
TTS_ENGINE=fish

# --- Fish Audio / SGLang (TTS_ENGINE=fish) ---
SGLANG_URL=http://10.8.0.3:8000
TTS_MODEL=fishaudio/drama-3
TTS_VOICE=default
# Optional voice-cloning reference (path on the Fish Audio server filesystem).
TTS_REFERENCE_AUDIO=/root/pipecat-voice/voice_profiles/solis.wav
# Optional transcript. If it looks like a .txt path, its contents are loaded;
# otherwise the raw string is used. Leave empty to auto-load <audio>.txt.
TTS_REFERENCE_TEXT=/root/pipecat-voice/voice_profiles/solis.txt

# --- Drama 3 de-emphasis (TTS_ENGINE=fish) ---
# Drama 3 can sound overly theatrical/recitative. Two knobs trim that register
# without changing the voice reference. Leave a sampling variable EMPTY to omit
# the field and fall back to the server default.
# Sampling temperature sent to SGLang. Lower values reduce over-articulation;
# the recitative "radio drama" register lives in the distribution tails, so a
# lower value makes the delivery flatter. Tunable range: 0.3–0.6.
TTS_TEMPERATURE=0.45
# Nucleus sampling top-p: trims the tail of the acoustic token distribution
# (exaggerated intonations). 0.7–0.8 is the useful range.
TTS_TOP_P=0.75
# Optional top-k. Leave empty to omit the field (server default).
TTS_TOP_K=
# Neutral style tag prepended to every synthesis request that does not already
# start with a "[...]" tag. LLM replies carry no tag (the system prompt forbids
# stage directions), so they get this register; clip banks with their own tag
# are left untouched. Empty disables the prefix.
TTS_STYLE_TAG="[conversational, casual]"

# --- Custom voice post-processing (/customvoice page) ---
# A freshly recorded reference is cleaned and level-matched before it is
# installed. Disable the whole chain to store the plain transcode.
CUSTOMVOICE_POSTPROCESS=true
# Noise reduction (FFT denoise) + de-essing. Turn off for a clean studio take.
CUSTOMVOICE_DENOISE=true
# Integrated loudness target (LUFS) and true-peak ceiling (dBTP). Every take
# lands on the same level, so recordings never jump in volume. Kept below the
# shipped reference (solis.wav measures ≈-23.4 LUFS): a user take is far more
# dynamic, so at equal LUFS its hotter peaks make the cloned voice sound louder.
CUSTOMVOICE_TARGET_LUFS=-26
CUSTOMVOICE_TRUE_PEAK=-5

# --- Kokoro (TTS_ENGINE=kokoro) ---
KOKORO_URL=http://localhost:8880
KOKORO_VOICE=af_heart

# --- STT (Parakeet TDT, pure-Rust ONNX pipeline) ---
# Directory containing nemo128.onnx, encoder-model*.onnx,
# decoder_joint-model.onnx, vocab.txt and config.json.
STT_MODEL_PATH=models/parakeet
# Optional external tokenizer; leave empty — Parakeet decodes text internally.
STT_TOKENIZER_PATH=

# --- VAD (Silero VAD v5, ONNX) ---
# Master switch. When disabled, the STT worker falls back to its internal
# RMS-based segmentation (legacy behaviour).
VAD_ENABLED=true
# Path to the Silero VAD ONNX model (16 kHz, 512-sample windows).
VAD_MODEL_PATH=models/silero_vad.onnx
# Speech probability above which a window counts as voice.
VAD_THRESHOLD=0.5
# Probability below which a window counts as silence. Must be <= VAD_THRESHOLD;
# the gap provides hysteresis and prevents onset/offset flapping.
VAD_NEG_THRESHOLD=0.35
# Speech required (ms) to confirm an utterance start.
VAD_MIN_SPEECH_MS=250
# Silence required (ms) to close an utterance (end-of-utterance window).
VAD_MIN_SILENCE_MS=700
# Audio kept (ms) before the detected onset, so the first phoneme is not clipped.
VAD_SPEECH_PAD_MS=30
# Speech required (ms) to fire the fast barge-in onset (must be <= VAD_MIN_SPEECH_MS).
VAD_BARGE_IN_MIN_SPEECH_MS=150
# Utterance gate applied before transcription. A VAD segment is dropped (no STT,
# no LLM turn) when it is shorter than VAD_MIN_UTTERANCE_MS ms (0 disables) or
# its average speech probability is below VAD_MIN_AVG_PROB (0.0 disables). The
# average is taken over the voiced windows only (prob >= VAD_NEG_THRESHOLD), so
# the min_silence hangover does not dilute short words ("uno", "due"). This
# filters the VAD false-triggers (breaths, clicks, acoustic echo, noise) that the
# STT turns into hallucinated text ("Bye.", "She", <transcript 0 tokens>) and
# that would each otherwise cost a full LLM turn. Raise the probability to be
# stricter; lower it if genuine soft speech is being dropped.
VAD_MIN_UTTERANCE_MS=300
VAD_MIN_AVG_PROB=0.25
# Feed the STT engine with VAD-segmented utterances (true) instead of letting
# the worker run its own RMS segmentation (false).
VAD_SEGMENT_STT=true
# Interrupt the agent when the user starts speaking over it.
VAD_BARGE_IN=true
# Log the speech probability of every analysis window (debug).
VAD_DEBUG=false

# --- Fillers / Mixer ---
# Directory of preloaded "*.raw" PCM clips (16-bit LE mono, 44.1 kHz), e.g.
# uhm / eh / mmm. Played round-robin at most once per turn.
FILLERS_DIR=assets/fillers
# Master switch. When disabled, no end-of-turn filler is ever queued.
FILLER_ENABLED=true
# Latency guard (ms): the filler is a latency mask, so it is only played if the
# response audio has not started within this window. If the reply starts
# speaking earlier the filler is dropped, instead of talking over the first
# words of the answer. Increase for slower LLM/TTS, decrease for faster ones.
FILLER_DELAY_MS=2500
FILLER_GAIN_IDLE=1.0
FILLER_GAIN_DUCK=0.15
# --- Start filler (turn-start acknowledgment when the reply is late) ---
# A second latency mask, on its own mixer channel: if the first audio chunk of
# the reply has not reached the mixer within START_FILLER_GO_DELAY ms, one of the
# phrases below (picked at random) is played so the agent acknowledges the turn
# while the LLM/TTS are still working. If audio arrives in time the cue is
# dropped. It shares the filler gain and is faded out (then cut) as soon as the
# reply starts playing, so it does not overlap the answer.
START_FILLER_ENABLED=true
# How long (ms) to wait for the first audio chunk of the reply before playing the
# start filler (default 1200). Measured from chunk ARRIVAL (not from playback
# start, which would also include the TTS_PREBUFFER_MS jitter buffer). Same idea
# as FILLER_DELAY_MS, but shorter. Raised above the default so the cue only fires
# when the first audio is genuinely late.
START_FILLER_GO_DELAY=1800
# Phrases spoken by the start filler, ONE PER LINE (a single line with "|"
# separators also works). One clip is synthesized per phrase and a RANDOM one is
# played when the reply is late, so the cue varies turn to turn. The bank follows
# this list: editing it triggers a full re-synthesis at the next restart (no
# manual deletion needed), phrases dropped here are deleted on disk, and a
# failed synthesis keeps the previous clip. An unchanged list is skipped, so
# startup stays fast. Drama-3 style tags ([calm], [sigh], ...) are passed
# through to the TTS.
START_FILLER_TEXTS="[calm] Un momento...
[thoughtful] A ver, veamos...
[hesitant] Entonces... echemos un vistazo...
[sigh] Vale, ya voy..."
# Directory holding the start-filler clips (kept separate from FILLERS_DIR so
# they are not added to the round-robin filler bank).
START_FILLER_DIR=assets/start_filler
# TTS jitter buffer (ms): how much synthesized audio the mixer buffers before it
# starts playing the reply. Lower values start speaking sooner; higher values
# absorb uneven TTS chunk arrival at the cost of added latency. Drama-3's vocoder
# emits audio in windows of codec frames (2048 samples = 46.4 ms each); with the
# recommended server setting (stride 4, followup 8, overlap 16, crossfade 512)
# the first window is 4 frames = 185.8 ms and the steady windows are 8 frames =
# 371.6 ms. The first arrives at the server TTFB (~0.38 s) and the second ~350 ms
# later, so ANY value in (186, 557] ms primes on the same second window (TTFA
# ~0.73 s); 550 ms takes the whole ~557 ms cushion for free. That cushion absorbs
# the drain when several TTS requests share one GPU and produce slower than real
# time (measured ~1.5x concurrent vs ~0.94x alone); without it every chunk seam
# starves the queue and pauses playback (a micro-gap at each sentence start).
# Below ~186 ms playback starts on the first window (TTFA ~= server TTFB) but
# leaves almost no cushion. Mid-reply underruns resume as soon as any audio is
# queued again (no refill).
TTS_PREBUFFER_MS=550
# Number of TTS requests kept in flight at once. Reply chunks are synthesized
# ahead of playback (and replayed in order) so each request's start-up latency
# hides behind the previous chunk's audio instead of leaving an audible gap.
# 1 restores strictly serial synthesis.
TTS_PIPELINE_DEPTH=3

# --- Backchannel (short acknowledgments while the user speaks) ---
# Master switch. When disabled, no during-utterance acknowledgment is played.
BACKCHANNEL_ENABLED=true
# Directory of the generated "*.raw" PCM clips (16-bit LE mono, 44.1 kHz). One
# clip is synthesized per phrase of BACKCHANNEL_TEXTS at startup.
BACKCHANNEL_DIR=assets/backchannel
# Phrases spoken by the backchannel, ONE PER LINE (a single line with "|"
# separators also works). One clip is synthesized per phrase and a RANDOM one is
# played, so the cue varies from turn to turn. The bank follows this list:
# editing it triggers a full re-synthesis at the next restart (no manual
# deletion needed), phrases dropped here are deleted on disk, and a failed
# synthesis keeps the previous clip. An unchanged list is skipped, so startup
# stays fast. Drama-3 style tags ([calm], ...) are passed through to the TTS.
#
# A line may start with an optional "category::" prefix (e.g. "agree::..."); the
# scheduler uses it to progress through the turn (an acknowledgment first,
# agreement later) and to avoid chaining two cues of the same kind. Lines
# without a prefix belong to the default "ack" category.
BACKCHANNEL_TEXTS="ack::[calm] Mm-hmm...
ack::[attentive] Ajá...
agree::[agreeing] Claro...
ack::[listening] Ya veo..."
# Mix gain applied to the backchannel channel (0..1). Kept low so the cue stays
# in the background.
BACKCHANNEL_GAIN=0.55
# Confirmed speech required (ms) before the first cue may be considered.
BACKCHANNEL_MIN_SPEECH_MS=1500
# Base spacing (ms) between two cues of the same turn; randomized by ±JITTER/2.
BACKCHANNEL_INTERVAL_MS=3000
# Timing jitter (ms): delays the first cue by 0..JITTER after MIN_SPEECH_MS and
# spreads the spacing, so cues don't land on a fixed cadence.
BACKCHANNEL_JITTER_MS=1200
# Micro-pause window (ms): the cue lands only while the user is in a short
# pause. The reaction DEPTH is drawn at random inside [MIN, MAX] per pause, so
# cues don't all land at exactly MIN. Below MIN the user is likely still
# mid-word; above MAX (capped below VAD_MIN_SILENCE_MS) the utterance is about
# to close.
BACKCHANNEL_PAUSE_MIN_MS=96
BACKCHANNEL_PAUSE_MAX_MS=260
# Probability (0..1) of cueing when a qualifying micro-pause occurs.
BACKCHANNEL_PROB=0.65
# Random per-cue volume scaling (± fraction) so cues don't sit at a fixed level.
BACKCHANNEL_VOLUME_JITTER=0.2
# Maximum cues emitted per user turn.
BACKCHANNEL_MAX_PER_TURN=3

# --- Multilingual clip banks (backchannel + filler + start filler) ---
# Every clip bank lives in a per-language subdirectory `<dir>/<lang>` named after
# the ISO 639-1 code (<dir>/<lang>/, e.g. assets/backchannel/es/), including the
# default language LANG_DEFAULT. When the user is detected speaking a different
# language, that language's banks are generated on demand into their own
# subdirectory and the session switches to them. Each language has its own
# directory, so regenerating one never removes the clips of another.
LANG_DEFAULT=es
# Detect the user's language with a dedicated LLM call (a short auxiliary
# request that returns only the ISO code) and switch the banks when it differs
# from the active one. Requires the LLM (LLM_ENABLED=true).
LANG_DETECT_ENABLED=true
# Minimum transcript length (chars) before a sample is sent to the LLM.
LANG_MIN_CHARS=12
# The first user turn is always checked; afterwards the language is re-checked
# once every N turns, so a mid-conversation switch is picked up without paying
# for a detection call on every turn.
LANG_REDETECT_EVERY=5
# Built-in phrase tables cover es/en/it/fr/de/pt. For any other language the
# phrase list is generated by the LLM; set false to keep only the tables.
LANG_LLM_FALLBACK_ENABLED=true

# --- Per-session audit log ---
# Writes one structured JSONL file per WebSocket session: a configuration
# snapshot at the start, every inbound control message and outbound event, the
# cue emissions, the language switches, the barge-ins / interrupts and an
# event-count summary at the end. Useful for post-mortem analysis and for
# tuning latency knobs (FILLER_DELAY_MS, START_FILLER_GO_DELAY, ...). One
# <id>.jsonl file per session, written under SESSION_AUDIT_DIR.
SESSION_AUDIT_ENABLED=true
SESSION_AUDIT_DIR=logs/sessions

# --- External audio source (URL / radio / YouTube) ---
# Injects an audio stream — a direct media URL, an Icecast/Shoutcast radio or a
# YouTube link — into the mixer's own "external" channel, mixed under the
# agent's voice. Demonstrated in the /mivocero client (URL box + Play/Stop +
# volume slider). Decoding is delegated to external tools:
#   * ffmpeg  — decodes direct URLs, radio and HLS streams (required);
#   * yt-dlp  — resolves YouTube URLs to a decodable stream (YouTube only).
# A barge-in (interrupt) stops the agent but leaves this background source
# playing; the external channel does not count as the agent speaking.
# Gain applied to the external channel (kept below 1 so it sits under the voice).
EXTERNAL_AUDIO_GAIN=0.6
# Override the decoder binaries if they are not on PATH.
MEDIA_FFMPEG_BIN=ffmpeg
MEDIA_YTDLP_BIN=yt-dlp
# Extra yt-dlp arguments (space-separated), e.g. to pass cookies or a JS
# runtime: "--cookies /path/cookies.txt". Leave empty normally.
MEDIA_YTDLP_ARGS=

# --- LLM reasoning (OpenAI-compatible chat-completions endpoint) ---
# When enabled, the user transcript is sent to the model and the reply is
# spoken by the TTS (sentence by sentence while streaming). When disabled the
# pipeline is a plain STT → TTS echo. The `/mivocero` client shows both the
# user transcript and the assistant reply in its chat box.
# Enabled automatically when both LLM_BASE_URL and LLM_MODEL are set.
LLM_ENABLED=true
# Base URL including the API version, e.g. a local vLLM / SGLang server.
LLM_BASE_URL=http://10.8.0.4:8000/v1
# Sent as `Authorization: Bearer <key>`. Local servers accept any value.
LLM_API_KEY=EMPTY
LLM_MODEL=Qwen/Qwen3.6-35B-A3B
LLM_MAX_TOKENS=1024
LLM_TEMPERATURE=0.7
# Stream tokens as they are generated (true → lower time-to-first-audio) or
# wait for the complete reply before speaking (false).
LLM_STREAM=true
# Rolling conversation window kept in memory (messages, system prompt excluded).
LLM_HISTORY_MESSAGES=16
# --- Chain-of-thought control (three independent mechanisms) ---
# Some reasoning models narrate their thinking instead of answering directly:
# Qwen3 on vLLM/SGLang writes it to a separate `reasoning_content` field (or
# inline think blocks) when a reasoning parser is enabled, but with no parser
# at all it leaks straight into `content` and would be read aloud by the TTS.
# 1) Server-side thinking mode. `false` sends
#    `chat_template_kwargs: {"enable_thinking": false}` (vLLM/SGLang, Qwen3).
#    Leave unset/omitted to not send the vendor-specific field at all.
LLM_ENABLE_THINKING=false
# 2) Drop the reasoning from the reply before it reaches the chat box / TTS:
#    ignores `reasoning_content` deltas and removes inline think blocks (also
#    stable when a marker is split across streamed deltas).
LLM_STRIP_REASONING=true
# 3) Append an explicit "answer directly, never narrate your reasoning"
#    directive to the system prompt (see `DIRECT_ANSWER_DIRECTIVE`).
LLM_DIRECT_ANSWERS=true
# Streaming chunking: characters the reply must hold before the FIRST chunk is
# handed to the TTS. Small by default so the agent starts speaking on the first
# few words instead of waiting for a whole sentence. `0` disables the early
# word-boundary flush (chunks then end on sentence punctuation only).
LLM_STREAM_FIRST_CHUNK_CHARS=20
# Same threshold for every chunk after the first. Larger chunks give better
# prosody and fewer TTS requests: each chunk is a separate request with its own
# start-up and its own audio seam, so a larger value cuts the number of seams
# (where the TTS queue can starve) per reply.
LLM_STREAM_CHUNK_CHARS=240
# Optional system prompt override. Leave unset for the built-in voice-assistant
# prompt (see `DEFAULT_LLM_SYSTEM_PROMPT` in src/config.rs).
#LLM_SYSTEM_PROMPT=

# --- Tool calling (configurable HTTP tools) ---
# Lets the model call external HTTP services (weather, home automation, a
# webhook, …) before answering. The deployment's vLLM server runs without a
# tool parser, so it returns the call as text in the Qwen native
# `<tool_call>` format: the engine parses it client-side, executes the HTTP
# request and hands the result back to the model (see src/tools.rs). The
# tool-call block is never spoken nor shown in the chat box.
# Enabled automatically when LLM_TOOLS_FILE points to a valid tool file.
LLM_TOOLS_ENABLED=false
# JSON file with the tool definitions (name, description, JSON schema, url,
# method, headers). See tools.example.json.
LLM_TOOLS_FILE=
# Maximum tool-call rounds per turn (call → execute → re-prompt). Bounds
# runaway loops; the first round is the model's initial reply.
LLM_TOOL_MAX_ITERATIONS=4
# Timeout of a single tool HTTP request, in seconds.
LLM_TOOL_TIMEOUT_SECS=15
# Tool responses are truncated to this many characters before being handed
# back to the model (keeps the context small).
LLM_TOOL_RESULT_MAX_CHARS=2000

# --- HTTPS (optional) ---
# Set BOTH to serve HTTPS/WSS instead of plain HTTP. A secure context is
# required by browsers to grant microphone access (getUserMedia) when the page
# is not opened on localhost. Generate a self-signed pair with:
#   ./server.sh cert
# Then open https://localhost:8787/ and accept the certificate warning.
TLS_CERT=certs/cert.pem
TLS_KEY=certs/key.pem