From fe5c23af3c46d6a6168d9b77bd703065aa65cbc5 Mon Sep 17 00:00:00 2001
From: cai <cai@nbcai.cc>
Date: Sat, 22 Aug 2026 10:32:18 +0800
Subject: [PATCH] feat(asr): propagate participant language per session
---
src/asr_realtime.rs | 31 +++++++++++++++++++++++++++++++
src/main.rs | 7 +++++++
2 files changed, 38 insertions(+), 0 deletions(-)
diff --git a/src/asr_realtime.rs b/src/asr_realtime.rs
index ad99ff2..3903ca0 100644
--- a/src/asr_realtime.rs
+++ b/src/asr_realtime.rs
@@ -34,6 +34,7 @@
pub(crate) struct AudioIngressMetadata {
pub(crate) input_source_category: String,
pub(crate) client_fixture_sequence: String,
+ pub(crate) language: Option<String>,
}
impl AudioIngressMetadata {
@@ -90,9 +91,16 @@
match (source, sequence) {
(None, None) => Ok(None),
(Some("controlled_fixture"), Some(sequence)) if valid_sequence(sequence) => {
+ let language = attributes.get("language").map(String::as_str);
+ if let Some(language) = language {
+ if !valid_language(language) {
+ return Err("invalid_language");
+ }
+ }
Ok(Some(Self {
input_source_category: "controlled_fixture".to_string(),
client_fixture_sequence: sequence.to_string(),
+ language: language.map(str::to_string),
}))
}
(Some(_), _) => Err("invalid_source_or_sequence"),
@@ -107,6 +115,10 @@
&& value
.chars()
.all(|ch| ch.is_ascii_alphanumeric() || matches!(ch, '-' | '_' | '.'))
+}
+
+fn valid_language(value: &str) -> bool {
+ matches!(value, "ja-JP" | "zh-CN")
}
impl RealtimeAsrConfig {
@@ -417,6 +429,9 @@
if let Some(metadata) = ingress_metadata {
line["inputSourceCategory"] = json!(metadata.input_source_category);
line["clientFixtureSequence"] = json!(metadata.client_fixture_sequence);
+ if let Some(language) = metadata.language.as_deref() {
+ line["language"] = json!(language);
+ }
}
line["audioIngressOriginStatus"] = json!(match ingress_metadata {
Some(_) => "controlled_fixture_bound",
@@ -546,12 +561,14 @@
"clientFixtureSequence".to_string(),
"fixture-01".to_string(),
);
+ attributes.insert("language".to_string(), "ja-JP".to_string());
let metadata =
AudioIngressMetadata::from_participant(&attributes).expect("valid attributes");
assert_eq!(
Some(AudioIngressMetadata {
input_source_category: "controlled_fixture".to_string(),
client_fixture_sequence: "fixture-01".to_string(),
+ language: Some("ja-JP".to_string()),
}),
metadata
);
@@ -589,6 +606,16 @@
"ordinary_mic_absent",
AudioIngressMetadata::origin_status(&HashMap::new())
);
+ attributes.insert(
+ "clientFixtureSequence".to_string(),
+ "fixture-01".to_string(),
+ );
+ attributes.insert("language".to_string(), "ja-jp".to_string());
+ assert_eq!(
+ Err("invalid_language"),
+ AudioIngressMetadata::from_participant(&attributes)
+ );
+ attributes.insert("language".to_string(), "ja-JP".to_string());
assert_eq!(
"participant_attributes_absent",
AudioIngressMetadata::rejection_status("incomplete_metadata")
@@ -625,6 +652,7 @@
let bound = AudioIngressMetadata {
input_source_category: "controlled_fixture".to_string(),
client_fixture_sequence: "fixture-01".to_string(),
+ language: Some("ja-JP".to_string()),
};
let absent = serde_json::from_slice::<serde_json::Value>(
&session_start_line("call-001", "trace-001", "turn-0001", "nonce-001", None)
@@ -646,6 +674,7 @@
.expect("bound json");
assert_eq!("controlled_fixture", with_metadata["inputSourceCategory"]);
assert_eq!("fixture-01", with_metadata["clientFixtureSequence"]);
+ assert_eq!("ja-JP", with_metadata["language"]);
assert_eq!(
"controlled_fixture_bound",
with_metadata["audioIngressOriginStatus"]
@@ -653,6 +682,7 @@
let next = AudioIngressMetadata {
input_source_category: "controlled_fixture".to_string(),
client_fixture_sequence: "fixture-02".to_string(),
+ language: Some("zh-CN".to_string()),
};
let next_line = session_start_line(
"call-001",
@@ -664,6 +694,7 @@
.expect("next bound session line");
let next_value: serde_json::Value = serde_json::from_slice(&next_line).expect("next json");
assert_eq!("fixture-02", next_value["clientFixtureSequence"]);
+ assert_eq!("zh-CN", next_value["language"]);
assert_ne!(
with_metadata["clientFixtureSequence"],
next_value["clientFixtureSequence"]
diff --git a/src/main.rs b/src/main.rs
index b92ad4b..0e179f5 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -5040,6 +5040,7 @@
"clientFixtureSequence".to_string(),
"fixture-01".to_string(),
),
+ ("language".to_string(), "ja-JP".to_string()),
]);
let mut starts = Vec::new();
for (session_index, sequence) in [(1, "fixture-01"), (2, "fixture-02")] {
@@ -5065,6 +5066,7 @@
let is_in_speech = vad.in_speech;
assert!(!was_in_speech && is_in_speech);
attributes.insert("clientFixtureSequence".to_string(), sequence.to_string());
+ attributes.insert("language".to_string(), "ja-JP".to_string());
let metadata = AudioIngressMetadata::from_participant(&attributes)
.expect("valid participant attributes")
.expect("controlled fixture metadata");
@@ -5079,6 +5081,7 @@
let session_json: serde_json::Value =
serde_json::from_slice(&session_line).expect("session start json");
assert_eq!(sequence, session_json["clientFixtureSequence"]);
+ assert_eq!("ja-JP", session_json["language"]);
starts.push(metadata.client_fixture_sequence);
vad.reset_current_turn();
}
@@ -5171,6 +5174,7 @@
"clientFixtureSequence".to_string(),
"fixture-01".to_string(),
),
+ ("language".to_string(), "ja-JP".to_string()),
]);
let mut upload = None;
let mut last_fixture_sequence = None;
@@ -5207,6 +5211,7 @@
"clientFixtureSequence".to_string(),
"fixture-02".to_string(),
);
+ attrs.insert("language".to_string(), "zh-CN".to_string());
let (was, is, turn) = observe_bound_participant_frame(
"participant-user",
Some("participant-user"),
@@ -5237,6 +5242,8 @@
.expect("second session request");
assert!(first_request.contains("\"clientFixtureSequence\":\"fixture-01\""));
assert!(second_request.contains("\"clientFixtureSequence\":\"fixture-02\""));
+ assert!(first_request.contains("\"language\":\"ja-JP\""));
+ assert!(second_request.contains("\"language\":\"zh-CN\""));
assert!(
first_request.contains("\"audioIngressOriginStatus\":\"controlled_fixture_bound\"")
);
--
Gitblit v1.9.3