Adding Realtime ASR Client

.circleci/config.yml

@@ -8,7 +8,7 @@ jobs:
     steps:
       - run:
           name: "Install build dependencies"
-          command: "sudo apt-get --allow-releaseinfo-change update && sudo apt-get install -y wget libasound2-dev libopus-dev libopusfile-dev"
+          command: "sudo apt-get --allow-releaseinfo-change update && sudo apt-get install -y wget libasound2-dev libopus-dev libopusfile-dev libboost-all-dev"
       - run:
           name: "Install bazel"
           command: "wget https://github.com/bazelbuild/bazelisk/releases/download/v1.11.0/bazelisk-linux-amd64 && sudo mv bazelisk-linux-amd64 /usr/local/bin/bazelisk && sudo chmod +x /usr/local/bin/bazelisk"

Dockerfile

@@ -10,7 +10,8 @@ RUN apt-get update && apt-get install -y \
     libasound2t64 \
     libogg0 \
     openssl \
-    ca-certificates
+    ca-certificates \
+    libboost-all-dev
 
 FROM base AS builddep
 ARG BAZEL_VERSION
@@ -67,4 +68,5 @@ COPY --from=builder /opt/riva/clients/nlp/riva_nlp_punct /usr/local/bin/
 COPY --from=builder /opt/riva/clients/nmt/riva_nmt_t2t_client /usr/local/bin/
 COPY --from=builder /opt/riva/clients/nmt/riva_nmt_streaming_s2t_client /usr/local/bin/
 COPY --from=builder /opt/riva/clients/nmt/riva_nmt_streaming_s2s_client /usr/local/bin/
+COPY --from=builder /opt/riva/clients/realtime/riva_realtime_asr_client /usr/local/bin/
 COPY examples /work/examples

README.md

@@ -8,6 +8,7 @@ NVIDIA Riva is a GPU-accelerated SDK for building Speech AI applications that ar
 - **Automatic Speech Recognition (ASR)**
     - `riva_streaming_asr_client`
     - `riva_asr_client`
+    - `riva_realtime_asr_client`
 - **Speech Synthesis (TTS)**
     - `riva_tts_client`
     - `riva_tts_perf_client`
@@ -73,6 +74,7 @@ You can find the built binaries in `bazel-bin/riva/clients`
 Riva comes with 2 ASR clients:
 1. `riva_asr_client` for offline usage. Using this client, the server will wait until it receives the full audio file before transcribing it and sending it back to the client.
 2. `riva_streaming_asr_client` for online usage. Using this client, the server will start transcribing after it receives a sufficient amount of audio data, "streaming" intermediate transcripts as it goes on back to the client. By default, it is set to transcribe after every `100ms`, this can be changed using the `--chunk_duration_ms` command line flag.
+3. `riva_realtime_asr_client` for realtime (websocket) usage. This client establishes a persistent websocket connection to the server, allowing for bidirectional real-time communication. The server will start transcribing after it receives a sufficient amount of audio data and continuously stream intermediate transcripts back to the client as it processes the audio. By default, it is set to transcribe after every `100ms`, which can be changed using the `--chunk_duration_ms` command line flag.
 
 To use the clients, simply pass in a folder containing audio files or an individual audio file name with the `audio_file` flag:
 ```
@@ -82,6 +84,10 @@ or
 ```
 $ riva_asr_client --audio_file audio_folder
 ```
+or
+```
+$ riva_realtime_asr_client --audio_file individual_audio_file.wav
+```
 
 Note that only single-channel audio files in the `.wav` format are currently supported.
 

WORKSPACE

@@ -102,3 +102,11 @@ http_archive(
     strip_prefix = "platforms-1.0.0",
     sha256 = "852b71bfa15712cec124e4a57179b6bc95d59fdf5052945f5d550e072501a769",
 )
+
+http_archive(
+    name = "websocketpp",
+    urls = ["https://github.com/zaphoyd/websocketpp/archive/refs/tags/0.8.2.tar.gz"],
+    sha256 = "6ce889d85ecdc2d8fa07408d6787e7352510750daa66b5ad44aacb47bea76755",
+    strip_prefix = "websocketpp-0.8.2",
+    build_file = "//third_party:BUILD.websocketpp"
+)

riva/clients/realtime/BUILD

@@ -0,0 +1,59 @@
+"""
+Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
+NVIDIA CORPORATION and its licensors retain all intellectual property
+and proprietary rights in and to this software, related documentation
+and any modifications thereto.  Any use, reproduction, disclosure or
+distribution of this software and related documentation without an express
+license agreement from NVIDIA CORPORATION is strictly prohibited.
+"""
+
+package(
+    default_visibility = ["//visibility:public"],
+)
+
+cc_library(
+    name = "realtime_audio_client_lib",
+    srcs = [
+        "audio_chunks.cpp",
+        "base_client.cpp", 
+        "realtime_client.cpp",
+    ],
+    hdrs = [
+        "audio_chunks.h",
+        "base_client.h",
+        "realtime_client.h",
+    ],
+    deps = [
+        "//riva/utils/wav:reader",
+        "//riva/utils/stats_builder:stats_builder_lib",
+        "@websocketpp//:websocketpp",
+        "@rapidjson//:rapidjson",
+        "@glog//:glog",
+        "@com_github_gflags_gflags//:gflags",
+    ],
+) 
+
+cc_binary(
+    name = "riva_realtime_asr_client",
+    srcs = ["riva_realtime_asr_client.cc"],
+    includes = ["-Irealtime"],
+    deps = [
+        ":realtime_audio_client_lib",
+        "@websocketpp//:websocketpp",
+        "@rapidjson//:rapidjson",
+        "//riva/utils/stats_builder:stats_builder_lib",
+        "//riva/utils/wav:reader",
+    ] + select({
+        "@platforms//cpu:aarch64": [
+            "@alsa_aarch64//:libasound"
+        ],
+        "//conditions:default": [
+            "@alsa//:libasound"
+        ],
+    }),
+    linkopts = [
+        "-lssl",
+        "-lcrypto",
+        "-lboost_system",
+    ]
+)