能力标签

🤖 Agent 🔄 工作流 🐳 Docker 🔗 REST API 🔊 TTS 🎙 STT 🧠 Claude

⚙️

Agent工作流

VoiceBlender 语音控制平台

基于 Go · 无代码搭建完整 AI 自动化流程

英文名：voiceblender

⭐ 68 Stars 🍴 8 Forks 💻 Go 📄 MIT 🏷 AI 8.2分

8.2AI 综合评分

语音AIWebRTC实时通信

⬇ 下载源码 ZIP ⚙️ 配置说明

✦ AI Skill Hub 推荐

AI Skill Hub 强烈推荐：VoiceBlender 语音控制平台是一款优质的Agent工作流。AI 综合评分 8.2 分，在同类工具中表现稳健。如果你正在寻找可靠的Agent工作流解决方案，这是一个值得深入了解的选择。

📚 深度解析

VoiceBlender 语音控制平台是一套完整的 AI Agent 自动化工作流方案。随着 AI 能力的不断提升，基于 Agent 的自动化工作流正在成为提升个人和团队效率的核心方式。区别于传统的 RPA 自动化（模拟鼠标键盘操作），AI Agent 工作流通过理解任务意图、动态规划执行路径，能够处理更复杂的非结构化任务。

VoiceBlender 语音控制平台工作流的设计遵循"最小配置，最大复用"原则：核心逻辑已经封装好，用户只需配置自己的 API Key 和业务参数即可快速上手。工作流内置错误处理和重试机制，在网络波动或 API 限速等情况下仍能稳定运行，适合作为生产环境的自动化基础设施。

在实际部署时，建议先在测试环境中运行 3-5 次，验证各个环节的输出结果符合预期，再部署到生产环境。AI Skill Hub 评分 8.2 分，是同类 Agent 工作流中的精选推荐。

📋 工具概览

一个可编程的开源语音平台，支持SIP和WebRTC通话控制及多方音频混音。它集成了ASR和音频处理能力，旨在为AI Agent提供实时语音交互基础设施，适合需要构建复杂语音工作流的开发者。

VoiceBlender 语音控制平台是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排，将复杂的多步骤任务拆解为清晰的自动化流程，实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成，适合构建数据处理管线、业务自动化和 AI 辅助决策系统。

GitHub Stars

⭐ 68

开发语言

支持平台

Windows / macOS / Linux（跨平台）

维护状态

轻量级项目，按需更新

开源协议

MIT

AI 综合评分

8.2 分

工具类型

Agent工作流

Forks

📖 中文文档

以下内容由 AI Skill Hub 根据项目信息自动整理，如需查看完整原始文档请访问底部「原始来源」。

📌 核心特色

可视化 Agent 工作流编排，无需编写复杂代码
支持多步骤自动化任务链，实现全流程无人值守
与外部 API、数据库和第三方服务无缝集成
内置错误处理与自动重试机制，保障稳定运行
提供可复用的自动化模板，快速在同类场景部署

🎯 主要使用场景

自动化日常重复性工作，将精力集中于创造性任务
构建数据采集 → 处理 → 输出的完整自动化管线
实现跨平台、跨系统的数据流转和业务协同

以下安装命令基于项目开发语言和类型自动生成，实际以官方 README 为准。

安装命令

# 方式一：go install（推荐）
go install github.com/VoiceBlender/voiceblender@latest

# 方式二：从源码编译
git clone https://github.com/VoiceBlender/voiceblender
cd voiceblender
go build -o voiceblender .

# 方式三：下载预编译二进制
# 访问 Releases 页面下载对应平台二进制文件
# https://github.com/VoiceBlender/voiceblender/releases

📋 安装步骤说明

访问 GitHub 仓库获取工作流文件
在对应平台（Dify / Flowise / Make 等）中找到「导入工作流」功能
上传工作流文件
按照提示配置必要的环境变量和 API Key
运行测试确认流程正常后投入使用

以下用法示例由 AI Skill Hub 整理，涵盖最常见的使用场景。

常用命令 / 代码示例

# 查看帮助
voiceblender --help

# 基本运行
voiceblender [options] <input>

# 详细使用说明请查阅文档
# https://github.com/VoiceBlender/voiceblender

以下配置示例基于典型使用场景生成，具体参数请参照官方文档调整。

配置示例

# voiceblender 配置说明
# 查看配置选项
voiceblender --config-example > config.yml

# 常见配置项
# output_dir: ./output
# log_level: info
# workers: 4

# 环境变量（覆盖配置文件）
export VOICEBLENDER_CONFIG="/path/to/config.yml"

📑 README 深度解析真实文档完整度 90/100 查看 GitHub 原文 →

以下内容由系统直接从 GitHub README 解析整理，保留代码块、表格与列表结构。

VoiceBlender

A Go service that bridges SIP and WebRTC voice calls with multi-party audio mixing, a REST API, and real-time webhooks.

Features

SIP inbound & outbound -- receive and originate SIP calls with codec negotiation (PCMU, PCMA, G.722, Opus), digest auth, session timers (RFC 4028)
SIP over TLS -- optional TLS transport on a second port alongside UDP, reusable by classic SIP trunks and required by WhatsApp
Early media -- SIP 183 Session Progress with SDP for pre-answer audio (custom ringback, IVR)
Hold/unhold -- SIP re-INVITE with sendonly/sendrecv direction
WebRTC -- browser-based voice via SDP offer/answer with trickle ICE
WhatsApp Business Calling -- inbound and outbound calls over SIP-TLS + ICE/DTLS-SRTP + Opus
WebSocket legs -- inbound (HTTP upgrade) and outbound (dial) PCM-over-WebSocket legs with binary or json_base64 framing, configurable sample rate (8/16/24/48 kHz), bidirectional text, and caller-supplied X-/P- headers — designed to also back a future generic Agent API
MoQ legs (experimental, PoC) -- inbound Media-over-QUIC legs over WebTransport/HTTP/3 with Opus framed one frame per MoQ Object (LOC-style). Tracks mengelbart/moqtransport (IETF draft-11); browser interop with draft-16 clients (moqtail, moq.dev) is not expected to work out of the box. Disabled by default; enable with MOQ_ENABLED=true + MOQ_TLS_CERT_FILE / MOQ_TLS_KEY_FILE
Multi-party rooms -- mix N participants with mixed-minus-self audio at a configurable sample rate (8 kHz, 16 kHz, or 48 kHz per room; default 16 kHz)
Room bridging -- join two rooms' mixers (same sample rate) with live-configurable direction (bidirectional, one-way each way, or parked); echo-free via mixed-minus-self
Audio routing matrix -- per-room role-based routing for asymmetric audio (barge-in / whisper / supervisor monitor). Tag legs with a free-form role and declare a matrix of who-hears-whom by role. Applied atomically at leg-join time so a supervisor cannot momentarily bleed into the customer's audio. See API.md.
WebSocket room access -- join rooms from any client over a WebSocket with base64 PCM frames
DTMF -- send and receive RFC 4733 telephone-events
Real-Time Text (RTT) -- ITU-T T.140 over RTP per RFC 4103 with RFC 2198 redundancy;
Recording -- stereo WAV recording per-leg or per-room, multi-channel per-participant tracks, pause/resume (writes silence to preserve timeline while sensitive data is exchanged), optional S3 upload
Playback -- stream WAV/MP3 audio or built-in telephone tones into legs or rooms
TTS -- text-to-speech into legs or rooms (ElevenLabs, Google Cloud, AWS Polly)
STT -- real-time speech-to-text with partial transcripts (ElevenLabs)
AI Agent -- attach a conversational AI agent to a leg or room (ElevenLabs, VAPI, Pipecat, Deepgram) with mid-session context injection
Answering Machine Detection (AMD) -- per-call analysis of outbound call audio to classify the answerer as human, machine, no-speech, or not-sure; optional voicemail beep detection via Goertzel frequency analysis
Webhooks -- real-time event delivery with HMAC-SHA256 signing and retry; typed event data with CDR-style leg.disconnected (disposition, timing, quality)
WebSocket event stream (VSI) -- GET /v1/vsi streams all events and accepts commands (mute, hold, DTMF, room management) over a single persistent WebSocket; filter by app_id regex for multi-tenant isolation
Prometheus metrics -- operational metrics exposed at GET /metrics (active legs/rooms, call durations, disconnect reasons, Go runtime). See API.md for the full metric reference. Profiling via go tool pprof is available at /debug/pprof/ when built with -tags pprof.

Capabilities

Inbound — Meta-originated INVITEs are auto-routed to a WhatsApp handler when the From URI host ends in meta.vc. The leg comes up in ringing, fires leg.ringing (leg_type: "whatsapp_in"), and waits for POST /v1/legs/{id}/answer. The 200 OK then carries the pre-gathered ICE/DTLS-SRTP answer.
Outbound — POST /v1/legs {"type":"whatsapp", ...} returns 201 immediately with the leg in ringing. ICE gathering, the digest 401/407 round-trip, and the SDP-answer apply happen asynchronously; outcome is signalled via leg.connected or leg.disconnected.
Audio — full-duplex Opus at 48 kHz with mixed-minus-self room participation, recording, TTS, STT, agent attachment, speaking detection, playback. The mixer auto-resamples between WhatsApp's 48 kHz and your room's configured rate.
DTMF — inbound RFC 4733 telephone-events are decoded and emitted as dtmf.received plus the standard cross-leg broadcast.
Webhooks + WebSocket events — leg.ringing / leg.connected / leg.disconnected / dtmf.received / speaking.started / speaking.stopped all carry leg_type set to whatsapp_in or whatsapp_out so multi-tenant filtering works as it does for SIP and WebRTC legs.

Integration tests (requires two SIP instances)

go test -tags integration -v -timeout 60s ./tests/integration/

Dependencies

Library	Description	Notes
[sipgo](https://github.com/emiago/sipgo)	SIP stack	Excellent SIP stack in go
[pion/webrtc](https://github.com/pion/webrtc)	WebRTC	Nothing is better than Pion
[go-chi](https://github.com/go-chi/chi)	HTTP router
[zaf/g711](https://github.com/zaf/g711)	G.711 codec
[gobwas/ws](https://github.com/gobwas/ws)	WebSocket
[go-audio/wav](https://github.com/go-audio/wav)	WAV encoding
[gopus](https://github.com/thesyncim/gopus)	Opus codec	Thanks Marcelo! (Claude and Codex too!)
[go-mp3](https://github.com/hajimehoshi/go-mp3)	MP3 decoder	Pure Go
[go-audio/audio](https://github.com/go-audio/audio)	Audio buffer types
[google/uuid](https://github.com/google/uuid)	UUID generation
[prometheus/client_golang](https://github.com/prometheus/client_golang)	Prometheus metrics
[aws-sdk-go-v2](https://github.com/aws/aws-sdk-go-v2)	AWS SDK (S3, Polly)
[cloud.google.com/go/texttospeech](https://cloud.google.com/go/docs/reference/cloud.google.com/go/texttospeech/latest)	Google Cloud TTS
[protobuf](https://github.com/protocolbuffers/protobuf-go)	Protocol Buffers	Pipecat agent
[x/sync](https://pkg.go.dev/golang.org/x/sync)	Concurrency utilities

Build and run

go build -o voiceblender ./cmd/voiceblender ./voiceblender

Quick Start

```bash

Examples

Example	Description
[`examples/call_handler.py`](examples/call_handler.py)	Python webhook listener for inbound SIP calls with room conferencing
[`examples/webrtc-client/`](examples/webrtc-client/)	Browser-based WebRTC voice client with room management and DTMF
[`examples/gen_test_wav.py`](examples/gen_test_wav.py)	Generate test WAV files for playback testing

Configuration

All configuration is via environment variables:

Variable	Default	Description
`INSTANCE_ID`	(auto-generated UUID)	Instance identifier, included in API responses and webhooks
`HTTP_ADDR`	`:8080`	REST API listen address
`SIP_BIND_IP`	`127.0.0.1`	IPv4 address advertised in SDP/Contact/Via headers (and used as the listen address when `SIP_LISTEN_IP` is empty). Set to `0.0.0.0` for v4 wildcard, `::` for dual-stack on Linux when `bindv6only=0`.
`SIP_LISTEN_IP`	(same as SIP_BIND_IP)	UDP socket bind IP. Accepts `127.0.0.1`, `0.0.0.0`, `::`, or any literal v4/v6 address.
`SIP_BIND_IPV6`	(empty = v4-only)	IPv6 address advertised in SDP/Contact/Via for IPv6 calls. Set this for IPv6-only or dual-stack deployments.
`SIP_LISTEN_IPV6`	(same as SIP_BIND_IPV6)	Optional separate IPv6 socket bind address (e.g. when running with both `0.0.0.0` and a specific v6 literal).
`SIP_PORT`	`5060`	SIP listen port (UDP)
`SIP_TLS_PORT`	(disabled)	SIP-over-TLS listen port (typically `5061`). When set, `SIP_TLS_CERT` and `SIP_TLS_KEY` must also be provided. Required for WhatsApp Business Calling integration.
`SIP_TLS_CERT`		Path to PEM-encoded TLS certificate (e.g. `fullchain.pem`). Meta rejects self-signed certs — use a CA-signed cert matching a public FQDN.
`SIP_TLS_KEY`		Path to PEM-encoded TLS private key (e.g. `privkey.pem`).
`SIP_DEBUG`	`false`	When `true`, log the full RFC 3261 wire form of every inbound and outbound SIP request and response. Very verbose — use only for troubleshooting.
`SIP_DOMAIN`	(falls back to advertised IP)	FQDN advertised in From, Contact and Via on all outbound SIP signalling (classic trunks and WhatsApp). Should match the SAN on `SIP_TLS_CERT` and any allowlist your carrier or Meta keeps.
`SIP_HOST`	`voiceblender`	SIP User-Agent name
`ICE_SERVERS`	`stun:stun.l.google.com:19302`	STUN/TURN URLs (comma-separated)
`RECORDING_DIR`	`/tmp/recordings`	Local recording output directory
`LOG_LEVEL`	`info`	Log level (`debug`, `info`, `warn`, `error`)
`WEBHOOK_URL`		Default webhook URL for inbound calls
`ELEVENLABS_API_KEY`		API key for ElevenLabs TTS, STT, and Agent
`VAPI_API_KEY`		API key for VAPI Agent provider
`DEEPGRAM_API_KEY`		API key for Deepgram STT and TTS
`AZURE_SPEECH_KEY`		Subscription key for Azure Cognitive Speech Services (TTS and STT)
`AZURE_SPEECH_REGION`	`eastus`	Azure region for Speech Services (e.g. `eastus`, `westeurope`)
`S3_BUCKET`		S3 bucket for recording uploads
`S3_REGION`	`us-east-1`	AWS region
`S3_ENDPOINT`		Custom S3 endpoint (MinIO, etc.)
`S3_PREFIX`		Key prefix for S3 objects
`TTS_CACHE_ENABLED`	`false`	Enable disk-backed TTS audio cache. Cached audio persists across restarts.
`TTS_CACHE_DIR`	`/tmp/tts_cache`	Directory for cached TTS audio files (used when `TTS_CACHE_ENABLED=true`)
`TTS_CACHE_INCLUDE_API_KEY`	`false`	Include API key in TTS cache key (set `true` if different keys map to different voice clones)
`RTP_PORT_MIN`	`10000`	Minimum UDP port for RTP/RTCP media
`RTP_PORT_MAX`	`20000`	Maximum UDP port for RTP/RTCP media
`SIP_JITTER_BUFFER_MS`	`0`	SIP ingress jitter buffer target delay in ms (0 = disabled passthrough). Applies to every SIP leg.
`SIP_JITTER_BUFFER_MAX_MS`	`300`	Max depth of the SIP ingress jitter buffer (ms); frames beyond this are dropped oldest-first.
`SIP_EXTERNAL_IP`	(empty)	Public IPv4 address for NAT/Docker deployments. When set, used in SIP Contact headers and SDP media (c=) lines instead of the auto-detected or bind IP. IPv6 has no equivalent: set `SIP_BIND_IPV6` directly to the address you want advertised.
`DEFAULT_SAMPLE_RATE`	`16000`	Default mixer sample rate (Hz) for new rooms when `sample_rate` is not specified. Allowed: `8000`, `16000`, `48000`.
`SIP_REFER_AUTO_DIAL`	`false`	Accept incoming SIP REFER requests and auto-dial the transferred call. Default-deny (toll-fraud risk). Outbound transfers via the REST API are unaffected.
`SIP_AUTO_RINGING`	`false`	Behavior change vs prior releases: previously the server always sent `180 Ringing` after `100 Trying`. The new default sends only `100 Trying`; the API caller drives ringing explicitly via `POST /v1/legs/{id}/ring`, `/early-media`, or `/answer`. Set to `true` to restore the legacy auto-180 behavior.
`SIP_USE_SOURCE_SOCKET`	`false`	When `true`, route SIP responses and in-dialog requests (BYE, re-INVITE, INFO, NOTIFY, REFER) back to the request's source UDP socket instead of the peer's `Contact` URI / Via sent-by. Enable when peers advertise unroutable addresses (e.g. private IPs in `Contact` from behind NAT, or Via sent-by hosts that don't resolve). Equivalent to sipgo's `DialogUA.RewriteContact` plus per-response `SetDestination(req.Source())`.
`SPEECH_DETECTION_ENABLED`	`false`	Emit `speaking.started` / `speaking.stopped` events for every connected leg by default. Per-call `speech_detection` on `POST /v1/legs` or `POST /v1/legs/{id}/answer` overrides this.
`VSI_EVENT_BUFFER_SIZE`	`256`	Per-client buffer (in events) on the `/v1/vsi` WebSocket. When the client consumes events slower than they're produced, the buffer fills and new events are dropped (with a warn log on the leading edge of each drop burst and at every 10× threshold; the next delivered event also includes an `events_dropped` notification to the client). Clamped to `[16, 1_000_000]`. Tuning: larger values absorb longer back-pressure spikes at the cost of higher peak memory per client (roughly the average JSON event size × buffer size, e.g. ~1 KB × 256 ≈ 256 KB per connection at the default) and longer end-to-end latency for buffered events when the client recovers. Increase only if you observe drops on legitimate slow-consumer scenarios you can't fix at the client.
`MOQ_ENABLED`	`false`	Enable the experimental MoQ (Media over QUIC) inbound leg endpoint at `CONNECT /v1/legs/moq` over WebTransport/HTTP/3. PoC quality: tracks IETF draft-11 via `mengelbart/moqtransport`, single MoQ session per leg, Opus framed one frame per MoQ Object (LOC-style). When enabled, both `MOQ_TLS_CERT_FILE` and `MOQ_TLS_KEY_FILE` must be set.
`MOQ_LISTEN_ADDR`	`:8443`	UDP address for the HTTP/3 listener that backs the MoQ leg. Independent of `HTTP_ADDR` — TCP/`:8080` and UDP/`:8443` can run side-by-side.
`MOQ_TLS_CERT_FILE`	_(none)_	Path to the TLS certificate used by the HTTP/3 listener. Required when `MOQ_ENABLED=true`.
`MOQ_TLS_KEY_FILE`	_(none)_	Path to the TLS private key used by the HTTP/3 listener. Required when `MOQ_ENABLED=true`.
`MOQ_OPUS_BITRATE`	`24000`	Target bitrate (bps) for the Opus encoder feeding the MoQ leg's `mix` track. Must be in `6000..510000`.

VoiceBlender configuration

Set these env vars before starting voiceblender:

Variable	Value
`SIP_TLS_PORT`	`5061`
`SIP_TLS_CERT`	path to `fullchain.pem` for your FQDN
`SIP_TLS_KEY`	path to `privkey.pem`
`SIP_DOMAIN`	the FQDN you registered with Meta (must match the cert SAN)

Make a test outbound call:

curl -X POST http://localhost:8080/v1/legs \
  -H 'Content-Type: application/json' \
  -d '{
    "type": "whatsapp",
    "to": "+447900000000",
    "from": "+441300000000",
    "auth": { "password": "<meta-issued-digest-password>" },
    "room_id": "wa-test"
  }'

The HTTP response returns immediately with the leg in ringing; subscribe to the webhook or /v1/vsi event stream to see leg.connected (or leg.disconnected with a reason if Meta rejects the INVITE).

API Overview

Full reference: API.md

Typical Workflow

1. Register a webhook        POST /v1/webhooks
2. Receive inbound call      --> webhook: leg.ringing {leg_id, from, to}
3. Answer                    POST /v1/legs/{id}/answer
4. Create a room             POST /v1/rooms
5. Add legs to room          POST /v1/rooms/{id}/legs
6. Attach AI agent           POST /v1/legs/{id}/agent
7. Start recording           POST /v1/legs/{id}/record
8. Hang up                   DELETE /v1/legs/{id}

Troubleshooting

403 SIP server X.X.X.X from INVITE does not match any SIP server configured for phone number ... — SIP_DOMAIN doesn't match what's registered with Meta. Set it to the FQDN, not the IP, and confirm via the GET /settings query above.
404 Not Found on outbound — usually means the recipient phone number isn't a valid WhatsApp user, or the destination URI is malformed. Confirm the digits in to are the actual user's E.164 number.
Call connects but Meta sends BYE after 20 s with Reason: ... not receiving any media for a long time — your audio path (RTP/UDP egress) is being dropped before reaching Meta. Check firewall rules for outbound UDP from the RTP_PORT_MIN–RTP_PORT_MAX range and that ICE-srflx candidates are correct.
DTLS handshake stalls — Meta's offer is setup:actpass + ice-lite, and they don't initiate DTLS. VoiceBlender forces setup:active automatically; if you see pcmedia: DTLS state state=connecting for >5 s, run with LOG_LEVEL=debug and inspect pion's DTLS scope for the actual error.
Set SIP_DEBUG=true to log the full RFC 3261 wire form of every SIP message, including the auth-bearing retry after the 401/407 challenge — that's the most useful diagnostic for any signalling-layer issue.