tts: default language to auto, NULL lang selects auto, reject NULL text

docs/ARCHITECTURE.md

@@ -498,7 +498,7 @@ Input:
 
 Optional:
   --format <fmt>          WAV output format: wav16, wav24, wav32 (default: wav16)
-  --lang <name>           Language label (default: english)
+  --lang <name>           Language label (default: auto)
   --instruct <str>        Style instruction. Required for VoiceDesign, optional for
                           CustomVoice, rejected for Base
   --speaker <name>        Speaker name (CustomVoice only)

examples/server.sh

@@ -4,4 +4,4 @@
 ./build/tts-server \
     --model models/qwen-talker-1.7b-base-Q8_0.gguf \
     --codec models/qwen-tokenizer-12hz-Q8_0.gguf \
-    --host 127.0.0.1 --port 8080 --lang English
+    --host 127.0.0.1 --port 8080 --lang auto

src/pipeline-tts.cpp

@@ -439,8 +439,12 @@ qt_status pipeline_tts_synthesize(PipelineTTS *                pt,
     TtsPerf perf = {};
     Timer   t_total;
 
+    // NULL lang selects automatic language: the prompt carries no
+    // language id and the model infers it from the text.
+    const char * lang = params->lang ? params->lang : "auto";
+
     Timer t_build;
-    if (!prompt_builder_build(pt, tok, params->text, params->lang, instruct, speaker, ref_spk_emb_ptr, ref_text,
+    if (!prompt_builder_build(pt, tok, params->text, lang, instruct, speaker, ref_spk_emb_ptr, ref_text,
                               ref_codes_T > 0 ? ref_codes.data() : NULL, ref_codes_T, &prompt)) {
         return QT_STATUS_GENERATE_FAILED;
     }

src/qwen.cpp

@@ -313,6 +313,14 @@ enum qt_status qt_synthesize(struct qt_context * q, const struct qt_tts_params *
         return QT_STATUS_INVALID_PARAMS;
     }
 
+    if (!params->text || !params->text[0]) {
+        qt_set_error("qt_synthesize: params->text is NULL or empty");
+        if (out) {
+            qt_audio_free(out);
+        }
+        return QT_STATUS_INVALID_PARAMS;
+    }
+
     // Mode validation. Mirrors the upstream Python which raises
     // ValueError when generate_voice_design is called on a non
     // voice_design model and the same shape applies to

src/qwen.h

@@ -197,8 +197,10 @@ QT_API void qt_log_set(qt_log_cb cb, void * user_data);
 struct qt_tts_params {
     int abi_version;
 
-    // Input text and language hint. lang accepts the upstream
-    // qwen3-tts language names ("english", "chinese", "auto", ...).
+    // Input text and language hint. text is required and non empty.
+    // lang accepts the upstream qwen3-tts language names ("english",
+    // "chinese", "auto", ...). NULL selects "auto": the prompt carries
+    // no language id and the model infers it from the text.
     // instruct is the style instruction string; required for
     // voice_design, optional for custom_voice, rejected for base.
     // speaker is the named speaker for custom_voice models, rejected

tools/qwen-tts.cpp

@@ -34,7 +34,7 @@ static void print_usage(const char * prog) {
             "  stdin                   Target text to synthesise. Read fully then synthesised in one shot.\n\n"
             "Optional:\n"
             "  --format <fmt>          WAV output format: wav16, wav24, wav32 (default: wav16)\n"
-            "  --lang <name>           Language label (default: english)\n"
+            "  --lang <name>           Language label (default: auto)\n"
             "  --instruct <str>        Style instruction. Required for VoiceDesign, optional for\n"
             "                          CustomVoice, rejected for Base\n"
             "  --speaker <name>        Speaker name (CustomVoice only)\n"
@@ -130,7 +130,7 @@ static bool read_text_file(const char * path, std::string & out) {
 
 static bool parse_args(int argc, char ** argv, Args & a) {
     a                        = {};
-    a.lang                   = "english";
+    a.lang                   = "auto";
     a.format                 = "wav16";
     a.max_new_tokens         = 2048;
     a.seed                   = -1;

tools/tts-server.cpp

@@ -21,7 +21,7 @@ static void print_usage(const char * prog) {
             "Optional:\n"
             "  --host <ip>             Listen address (default: 127.0.0.1)\n"
             "  --port <n>              Listen port (default: 8080)\n"
-            "  --lang <name>           Language label (default: english)\n"
+            "  --lang <name>           Language label (default: auto)\n"
             "  --no-fa                 Disable flash attention\n"
             "  --clamp-fp16            Clamp hidden states to FP16 range\n",
             prog);
@@ -37,7 +37,7 @@ static std::string basename_of(const char * path) {
 int main(int argc, char ** argv) {
     const char *  talker_path = NULL;
     const char *  codec_path  = NULL;
-    std::string   lang        = "english";
+    std::string   lang        = "auto";
     server_config cfg;
     bool          use_fa     = true;
     bool          clamp_fp16 = false;